Public link to this notebook: https://colab.research.google.com/drive/1SOdNcVb4lfbeJTFfxs4HnFOf2GY_D-l4?usp=sharing .

New generalization split v_dat_p2_pp_moved_to_recipient predicted to be challenging for Wu et al 2023 baseline Transformers.¶

split out from https://colab.research.google.com/drive/1rvVNQYH7NUrLmsCfdcyzwMos-HMkCNTM#scrollTo=9qtNY-eruQ5r so can run in parallel

possible_issue_with_subj_pp_generalization_by_transformers_could_be_simple_nontree_pp_np_distractor_when_modifying_nps_with_related_nps_to_right.png

train ReCOGS baseline Transformer (from Wu et al 2023) on Wu's training set which only has v_dat_p2 pp modification on the theme (right most np), test generalization on recipient modification (left in np pair both right of verb)¶

we use a large machine here as we are testing the ReCOGS paper author's model (NOT the RASP model) with modified data

should be worse performance on the modified v_dat_p2 split at baseline (this is inverse of experiment above, where we expect an improvement by adding this data to the training set)

need to edit run_cogs.py to make it print out this score:

for k, v in per_cat_eval.items():
                if k  == "pp_recursion":
                    struct_pp_acc = 100 * v[0]/v[1]
                elif k  == "cp_recursion":
                    struct_cp_acc = 100 * v[0]/v[1]
                elif k  == "obj_pp_to_subj_pp":
                    struct_obj_subj_acc = 100 * v[0]/v[1]
                elif k  == "subj_to_obj_proper":
                    subj_to_obj_proper_acc = 100 * v[0]/v[1]
                elif k  == "prim_to_obj_proper":
                    prim_to_obj_proper_acc = 100 * v[0]/v[1]
                elif k  == "prim_to_subj_proper":
                    prim_to_subj_proper_acc = 100 * v[0]/v[1]
                elif k == "v_dat_p2_pp_moved_to_recipient":
                  v_dat_p2_pp_moved_to_recipient_acc = 100 * v[0]/v[1]
                else:
                    lex_acc += v[0]
                    lex_count += v[1]
            lex_acc /= lex_count
            lex_acc *= 100
            current_acc *= 100

            print(f"obj_pp_to_subj_pp: {struct_obj_subj_acc}")
            print(f"cp_recursion: {struct_cp_acc}")
            print(f"pp_recursion: {struct_pp_acc}")
            print(f"subj_to_obj_proper: {subj_to_obj_proper_acc}")
            print(f"prim_to_obj_proper: {prim_to_obj_proper_acc}")
            print(f"prim_to_subj_proper: {prim_to_subj_proper_acc}")
            print(f"v_dat_p2_pp_moved_to_recipient: {v_dat_p2_pp_moved_to_recipient_acc}")
            print(f"LEX: {lex_acc}")
            print(f"OVERALL: {current_acc}")

            results[f"{seed}_{data_variant}_{lf}"] = {
                "obj_pp_to_subj_pp" : struct_obj_subj_acc,
                "cp_recursion" : struct_cp_acc,
                "pp_recursion" : struct_pp_acc,
                "subj_to_obj_proper" : subj_to_obj_proper_acc,
                "prim_to_obj_proper" : prim_to_obj_proper_acc,
                "prim_to_subj_proper" : prim_to_subj_proper_acc,
                "v_dat_p2_pp_moved_to_recipient": v_dat_p2_pp_moved_to_recipient_acc,
                "lex_acc" : lex_acc,
                "overall_acc" : current_acc,
                "test_acc" : test_acc
            }

upload modified_train_set_examples_v_dat_p2_pp_moved_to_recipient.tsv from https://colab.research.google.com/drive/1IDs0EwIMp2wtLHk4KqnuGhuT3G14QEG1

In [ ]:
%cd /content/
!rm -rf ReCOGS
!git clone https://github.com/frankaging/ReCOGS.git
%cd ReCOGS
!mkdir cogs_train_default_wu_data_gen_on_modified_v_dat_p2

!cp recogs_positional_index/dev.tsv  cogs_train_default_wu_data_gen_on_modified_v_dat_p2/dev.tsv
!cp recogs_positional_index/train.tsv  cogs_train_default_wu_data_gen_on_modified_v_dat_p2/train.tsv

!cp recogs_positional_index/test.tsv cogs_train_default_wu_data_gen_on_modified_v_dat_p2/test.tsv
!cp recogs_positional_index/gen.tsv cogs_train_default_wu_data_gen_on_modified_v_dat_p2/gen.tsv

# we did the experiment in one direction that adding this to the training data
# now we also see how a fresh model does on this data after training on original training set
# training examples here are from https://colab.research.google.com/drive/1IDs0EwIMp2wtLHk4KqnuGhuT3G14QEG1
# each row is a row from Wu et al 2023's training data with rearrangement, no contamination of elsewhere or changing of vocabulary statistics
!tail +2 /content/modified_train_set_examples_v_dat_p2_pp_moved_to_recipient.tsv | sed -E 's/in_distribution/v_dat_p2_pp_moved_to_recipient/g' >> /content/gen_suffix.tsv
!head /content/gen_suffix.tsv
/content
Cloning into 'ReCOGS'...
remote: Enumerating objects: 436, done.
remote: Counting objects: 100% (124/124), done.
remote: Compressing objects: 100% (51/51), done.
remote: Total 436 (delta 96), reused 92 (delta 73), pack-reused 312 (from 1)
Receiving objects: 100% (436/436), 84.71 MiB | 36.94 MiB/s, done.
Resolving deltas: 100% (303/303), done.
Updating files: 100% (137/137), done.
/content/ReCOGS
Liam gave the monkey in the container a chalk .	Liam ( 0 ) ; * monkey ( 3 ) ; * container ( 6 ) ; chalk ( 8 ) ; give ( 1 ) AND agent ( 1 , 0 ) AND recipient ( 1 , 3 ) AND theme ( 1 , 8 ) AND nmod . in ( 3 , 6 )	v_dat_p2_pp_moved_to_recipient
Emma gave a landlord in a house the box .	Emma ( 0 ) ; landlord ( 3 ) ; house ( 6 ) ; * box ( 8 ) ; give ( 1 ) AND agent ( 1 , 0 ) AND recipient ( 1 , 3 ) AND theme ( 1 , 8 ) AND nmod . in ( 3 , 6 )	v_dat_p2_pp_moved_to_recipient
Emma awarded a bird on the stool the drink .	Emma ( 0 ) ; bird ( 3 ) ; * stool ( 6 ) ; * drink ( 8 ) ; award ( 1 ) AND agent ( 1 , 0 ) AND recipient ( 1 , 3 ) AND theme ( 1 , 8 ) AND nmod . on ( 3 , 6 )	v_dat_p2_pp_moved_to_recipient
Emma offered a girl on the table a drink .	Emma ( 0 ) ; girl ( 3 ) ; * table ( 6 ) ; drink ( 8 ) ; offer ( 1 ) AND agent ( 1 , 0 ) AND recipient ( 1 , 3 ) AND theme ( 1 , 8 ) AND nmod . on ( 3 , 6 )	v_dat_p2_pp_moved_to_recipient
Emma offered a teacher beside a bed the scarf .	Emma ( 0 ) ; teacher ( 3 ) ; bed ( 6 ) ; * scarf ( 8 ) ; offer ( 1 ) AND agent ( 1 , 0 ) AND recipient ( 1 , 3 ) AND theme ( 1 , 8 ) AND nmod . beside ( 3 , 6 )	v_dat_p2_pp_moved_to_recipient
Emma gave a mouse on a table the cake .	Emma ( 0 ) ; mouse ( 3 ) ; table ( 6 ) ; * cake ( 8 ) ; give ( 1 ) AND agent ( 1 , 0 ) AND recipient ( 1 , 3 ) AND theme ( 1 , 8 ) AND nmod . on ( 3 , 6 )	v_dat_p2_pp_moved_to_recipient
Emma lended a boy on a road the crystal .	Emma ( 0 ) ; boy ( 3 ) ; road ( 6 ) ; * crystal ( 8 ) ; lend ( 1 ) AND agent ( 1 , 0 ) AND recipient ( 1 , 3 ) AND theme ( 1 , 8 ) AND nmod . on ( 3 , 6 )	v_dat_p2_pp_moved_to_recipient
Mila gave the girl in a pit the cake .	Mila ( 0 ) ; * girl ( 3 ) ; pit ( 6 ) ; * cake ( 8 ) ; give ( 1 ) AND agent ( 1 , 0 ) AND recipient ( 1 , 3 ) AND theme ( 1 , 8 ) AND nmod . in ( 3 , 6 )	v_dat_p2_pp_moved_to_recipient
Elizabeth lended the girl in the bin a sandwich .	Elizabeth ( 0 ) ; * girl ( 3 ) ; * bin ( 6 ) ; sandwich ( 8 ) ; lend ( 1 ) AND agent ( 1 , 0 ) AND recipient ( 1 , 3 ) AND theme ( 1 , 8 ) AND nmod . in ( 3 , 6 )	v_dat_p2_pp_moved_to_recipient
The boy gave the girl on a boat a crystal .	* boy ( 1 ) ; * girl ( 4 ) ; boat ( 7 ) ; crystal ( 9 ) ; give ( 2 ) AND agent ( 2 , 1 ) AND recipient ( 2 , 4 ) AND theme ( 2 , 9 ) AND nmod . on ( 4 , 7 )	v_dat_p2_pp_moved_to_recipient
In [ ]:
!cat /content/gen_suffix.tsv >> cogs_train_default_wu_data_gen_on_modified_v_dat_p2/gen.tsv
In [ ]:
!pwd
/content/ReCOGS
In [ ]:
!ls cogs_train_default_wu_data_gen_on_modified_v_dat_p2
dev.tsv  gen.tsv  test.tsv  train.tsv

We just want to see how the performance is for the v_dat_p2_pp_moved_to_recipient split. It is expected to be bad (statistically worse than overall accuracy; I may also split out v_dat_p2_theme_modified later), similar to obj_pp_to_subj_pp , since it is out of distribution and similar mechanism.

In [ ]:
# Wu et al 2023 transformer with Wu et al 2023 baseline training data (no changes from me)
# How does it do on `np v_dat_p2 np pp np np` examples when it per the author's train.tsv it trains on `np v_dat_p2 np np pp np` ?
!python run_cogs.py --model_name ende_transformer --use_iiem --gpu 1 --train_batch_size 128 --eval_batch_size 128 --lr 0.0001 --data_path ./cogs_train_default_wu_data_gen_on_modified_v_dat_p2 --output_dir ./results_cogs_train_default_wu_data_gen_on_modified_v_dat_p2 --lfs cogs --do_train --do_test --do_gen --max_seq_len 512 --output_json --epochs 300 --seeds "42;66;77;88;99" # paper's seeds, not sure how they chose them
INFO:root:Baselining the Transformer Encoder-Decoder Model
INFO:root:__Number CUDA Devices: 1
INFO:root:Number of model params: 4344077
INFO:root:OUTPUT DIR: ./results_cogs_train_default_wu_data_gen_on_modified_v_dat_p2/cogs_pipeline.model.ende_transformer.lf.cogs.glove.False.seed.42
Epoch: 0:   0% 0/213 [00:00<?, ?it/s]We strongly recommend passing in an `attention_mask` since your input_ids may be padded. See https://huggingface.co/docs/transformers/troubleshooting#incorrect-output-when-padding-tokens-arent-masked.
/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 0: 100% 213/213 [00:11<00:00, 18.61it/s, loss=5.89]
Epoch: 1:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 1: 100% 213/213 [00:10<00:00, 20.25it/s, loss=4.54]
Epoch: 2:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 2: 100% 213/213 [00:10<00:00, 20.28it/s, loss=3.54]
Epoch: 3:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 3: 100% 213/213 [00:10<00:00, 20.26it/s, loss=2.49]
Epoch: 4:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 4: 100% 213/213 [00:10<00:00, 20.27it/s, loss=1.91]
Epoch: 5:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 5: 100% 213/213 [00:10<00:00, 20.26it/s, loss=1.58]
Epoch: 6:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 6: 100% 213/213 [00:10<00:00, 20.24it/s, loss=1.3]
Epoch: 7:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 7: 100% 213/213 [00:10<00:00, 20.25it/s, loss=1.09]
Epoch: 8:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 8: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.97]
Epoch: 9:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 9: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.87]
Epoch: 10:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 10: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.77]
Epoch: 11:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 11: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.68]
Epoch: 12:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 12: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.61]
Epoch: 13:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 13: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.54]
Epoch: 14:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 14: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.47]
Epoch: 15:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 15: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.4]
Epoch: 16:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 16: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.34]
Epoch: 17:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 17: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.29]
Epoch: 18:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 18: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.25]
Epoch: 19:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 19: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.22]
Epoch: 20:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 20: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.19]
Epoch: 21:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 21: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.17]
Epoch: 22:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 22: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.16]
Epoch: 23:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 23: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.14]
Epoch: 24:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 24: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.12]
Epoch: 25:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 25: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.12]
Epoch: 26:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 26: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.11]
Epoch: 27:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 27: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.1]
Epoch: 28:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 28: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.1]
Epoch: 29:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 29: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.08]
Epoch: 30:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 30: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.08]
Epoch: 31:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 31: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.07]
Epoch: 32:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 32: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.07]
Epoch: 33:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 33: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.06]
Epoch: 34:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 34: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.05]
Epoch: 35:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 35: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0.05]
Epoch: 36:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 36: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.05]
Epoch: 37:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 37: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.04]
Epoch: 38:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 38: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.04]
Epoch: 39:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 39: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.04]
Epoch: 40:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 40: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.04]
Epoch: 41:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 41: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.03]
Epoch: 42:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 42: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.03]
Epoch: 43:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 43: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.03]
Epoch: 44:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 44: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.04]
Epoch: 45:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 45: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.02]
Epoch: 46:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 46: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.02]
Epoch: 47:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 47: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.02]
Epoch: 48:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 48: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.02]
Epoch: 49:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 49: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.02]
Epoch: 50:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 50: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.02]
Epoch: 51:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 51: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.02]
Epoch: 52:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 52: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0.01]
Epoch: 53:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 53: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 54:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 54: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.02]
Epoch: 55:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 55: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.02]
Epoch: 56:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 56: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.02]
Epoch: 57:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 57: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.01]
Epoch: 58:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 58: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0.01]
Epoch: 59:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 59: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.01]
Epoch: 60:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 60: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.01]
Epoch: 61:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 61: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.01]
Epoch: 62:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 62: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.01]
Epoch: 63:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 63: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.01]
Epoch: 64:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 64: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.01]
Epoch: 65:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 65: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.01]
Epoch: 66:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 66: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.01]
Epoch: 67:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 67: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 68:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 68: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 69:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 69: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.01]
Epoch: 70:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 70: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.01]
Epoch: 71:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 71: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 72:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 72: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.01]
Epoch: 73:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 73: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 74:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 74: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.01]
Epoch: 75:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 75: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0.01]
Epoch: 76:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 76: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 77:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 77: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.01]
Epoch: 78:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 78: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.01]
Epoch: 79:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 79: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 80:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 80: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.01]
Epoch: 81:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 81: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 82:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 82: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 83:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 83: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 84:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 84: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 85:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 85: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 86:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 86: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 87:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 87: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.01]
Epoch: 88:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 88: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.01]
Epoch: 89:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 89: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.01]
Epoch: 90:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 90: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 91:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 91: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.01]
Epoch: 92:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 92: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 93:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 93: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 94:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 94: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 95:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 95: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 96:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 96: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 97:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 97: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 98:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 98: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 99:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 99: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 100:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 100: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 101:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 101: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 102:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 102: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 103:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 103: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 104:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 104: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 105:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 105: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 106:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 106: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 107:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 107: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 108:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 108: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 109:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 109: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 110:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 110: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 111:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 111: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 112:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 112: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 113:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 113: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 114:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 114: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 115:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 115: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 116:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 116: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 117:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 117: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 118:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 118: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 119:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 119: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 120:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 120: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 121:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 121: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 122:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 122: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 123:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 123: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 124:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 124: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 125:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 125: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 126:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 126: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 127:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 127: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 128:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 128: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 129:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 129: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 130:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 130: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 131:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 131: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 132:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 132: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 133:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 133: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 134:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 134: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 135:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 135: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 136:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 136: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 137:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 137: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 138:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 138: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 139:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 139: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 140:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 140: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 141:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 141: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 142:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 142: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 143:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 143: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 144:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 144: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 145:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 145: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 146:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 146: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 147:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 147: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 148:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 148: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 149:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 149: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 150:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 150: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 151:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 151: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 152:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 152: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 153:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 153: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 154:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 154: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 155:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 155: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 156:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 156: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 157:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 157: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 158:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 158: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 159:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 159: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 160:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 160: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 161:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 161: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 162:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 162: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 163:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 163: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 164:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 164: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 165:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 165: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 166:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 166: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 167:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 167: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 168:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 168: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 169:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 169: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 170:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 170: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 171:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 171: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 172:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 172: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 173:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 173: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 174:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 174: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 175:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 175: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 176:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 176: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 177:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 177: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 178:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 178: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 179:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 179: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 180:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 180: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 181:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 181: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 182:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 182: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 183:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 183: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 184:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 184: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 185:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 185: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 186:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 186: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 187:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 187: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 188:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 188: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 189:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 189: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 190:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 190: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 191:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 191: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 192:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 192: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 193:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 193: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 194:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 194: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 195:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 195: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 196:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 196: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 197:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 197: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 198:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 198: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 199:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 199: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 200:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 200: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 201:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 201: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 202:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 202: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 203:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 203: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 204:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 204: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 205:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 205: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 206:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 206: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 207:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 207: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 208:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 208: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 209:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 209: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 210:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 210: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 211:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 211: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 212:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 212: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 213:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 213: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 214:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 214: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 215:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 215: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 216:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 216: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 217:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 217: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 218:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 218: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 219:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 219: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 220:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 220: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 221:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 221: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 222:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 222: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 223:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 223: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 224:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 224: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 225:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 225: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 226:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 226: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 227:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 227: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 228:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 228: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 229:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 229: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 230:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 230: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 231:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 231: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 232:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 232: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 233:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 233: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 234:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 234: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 235:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 235: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 236:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 236: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 237:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 237: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 238:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 238: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 239:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 239: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 240:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 240: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 241:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 241: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 242:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 242: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 243:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 243: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 244:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 244: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 245:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 245: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 246:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 246: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 247:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 247: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 248:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 248: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 249:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 249: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 250:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 250: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 251:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 251: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 252:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 252: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 253:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 253: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 254:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 254: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 255:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 255: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 256:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 256: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 257:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 257: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 258:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 258: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 259:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 259: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 260:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 260: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 261:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 261: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 262:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 262: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 263:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 263: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 264:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 264: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 265:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 265: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 266:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 266: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 267:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 267: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 268:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 268: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 269:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 269: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 270:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 270: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 271:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 271: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 272:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 272: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 273:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 273: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 274:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 274: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 275:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 275: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 276:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 276: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 277:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 277: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 278:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 278: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 279:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 279: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 280:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 280: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 281:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 281: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 282:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 282: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 283:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 283: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 284:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 284: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 285:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 285: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 286:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 286: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 287:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 287: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 288:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 288: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 289:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 289: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 290:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 290: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 291:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 291: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 292:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 292: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 293:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 293: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 294:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 294: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 295:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 295: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 296:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 296: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 297:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 297: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 298:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 298: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 299:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 299: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 100% 300/300 [52:55<00:00, 10.58s/it]
INFO:root:Training is finished ...
Iteration: 100% 24/24 [00:10<00:00,  2.31it/s, acc=1]
Iteration: 100% 172/172 [20:51<00:00,  7.28s/it, acc=0.854]
obj_pp_to_subj_pp: 14.8
cp_recursion: 52.2
pp_recursion: 43.9
subj_to_obj_proper: 95.5
prim_to_obj_proper: 95.5
prim_to_subj_proper: 99.9
v_dat_p2_pp_moved_to_recipient: 43.292682926829265
LEX: 95.52000000000001
OVERALL: 85.38937409024744
INFO:root:Baselining the Transformer Encoder-Decoder Model
INFO:root:__Number CUDA Devices: 1
INFO:root:Number of model params: 4344077
INFO:root:OUTPUT DIR: ./results_cogs_train_default_wu_data_gen_on_modified_v_dat_p2/cogs_pipeline.model.ende_transformer.lf.cogs.glove.False.seed.66
Epoch: 0:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 0: 100% 213/213 [00:10<00:00, 20.22it/s, loss=5.99]
Epoch: 1:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 1: 100% 213/213 [00:10<00:00, 20.23it/s, loss=4.61]
Epoch: 2:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 2: 100% 213/213 [00:10<00:00, 20.24it/s, loss=3.61]
Epoch: 3:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 3: 100% 213/213 [00:10<00:00, 20.27it/s, loss=2.53]
Epoch: 4:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 4: 100% 213/213 [00:10<00:00, 20.22it/s, loss=1.96]
Epoch: 5:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 5: 100% 213/213 [00:10<00:00, 20.27it/s, loss=1.6]
Epoch: 6:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 6: 100% 213/213 [00:10<00:00, 20.28it/s, loss=1.32]
Epoch: 7:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 7: 100% 213/213 [00:10<00:00, 20.23it/s, loss=1.13]
Epoch: 8:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 8: 100% 213/213 [00:10<00:00, 20.27it/s, loss=1]
Epoch: 9:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 9: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.9]
Epoch: 10:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 10: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0.8]
Epoch: 11:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 11: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.71]
Epoch: 12:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 12: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.63]
Epoch: 13:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 13: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.54]
Epoch: 14:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 14: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.47]
Epoch: 15:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 15: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.41]
Epoch: 16:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 16: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.36]
Epoch: 17:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 17: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.31]
Epoch: 18:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 18: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.27]
Epoch: 19:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 19: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.23]
Epoch: 20:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 20: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.21]
Epoch: 21:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 21: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.17]
Epoch: 22:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 22: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.15]
Epoch: 23:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 23: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.13]
Epoch: 24:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 24: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.12]
Epoch: 25:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 25: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.11]
Epoch: 26:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 26: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.11]
Epoch: 27:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 27: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.09]
Epoch: 28:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 28: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0.09]
Epoch: 29:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 29: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.08]
Epoch: 30:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 30: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.07]
Epoch: 31:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 31: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.06]
Epoch: 32:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 32: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.05]
Epoch: 33:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 33: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0.05]
Epoch: 34:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 34: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.05]
Epoch: 35:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 35: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0.04]
Epoch: 36:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 36: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.04]
Epoch: 37:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 37: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.04]
Epoch: 38:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 38: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.04]
Epoch: 39:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 39: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.03]
Epoch: 40:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 40: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.03]
Epoch: 41:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 41: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.03]
Epoch: 42:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 42: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.03]
Epoch: 43:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 43: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.03]
Epoch: 44:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 44: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.02]
Epoch: 45:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 45: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.02]
Epoch: 46:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 46: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.02]
Epoch: 47:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 47: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.02]
Epoch: 48:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 48: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.02]
Epoch: 49:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 49: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.02]
Epoch: 50:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 50: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.02]
Epoch: 51:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 51: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.03]
Epoch: 52:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 52: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.02]
Epoch: 53:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 53: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.01]
Epoch: 54:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 54: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.02]
Epoch: 55:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 55: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.01]
Epoch: 56:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 56: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 57:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 57: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.01]
Epoch: 58:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 58: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.01]
Epoch: 59:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 59: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 60:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 60: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 61:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 61: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 62:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 62: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.01]
Epoch: 63:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 63: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.01]
Epoch: 64:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 64: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.01]
Epoch: 65:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 65: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.01]
Epoch: 66:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 66: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.01]
Epoch: 67:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 67: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.01]
Epoch: 68:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 68: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.01]
Epoch: 69:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 69: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 70:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 70: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 71:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 71: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.01]
Epoch: 72:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 72: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.01]
Epoch: 73:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 73: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.01]
Epoch: 74:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 74: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.01]
Epoch: 75:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 75: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.01]
Epoch: 76:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 76: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.01]
Epoch: 77:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 77: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 78:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 78: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 79:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 79: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 80:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 80: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 81:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 81: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.01]
Epoch: 82:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 82: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 83:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 83: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.01]
Epoch: 84:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 84: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 85:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 85: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 86:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 86: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0.01]
Epoch: 87:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 87: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 88:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 88: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.01]
Epoch: 89:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 89: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 90:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 90: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 91:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 91: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 92:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 92: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 93:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 93: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 94:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 94: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 95:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 95: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.01]
Epoch: 96:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 96: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 97:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 97: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 98:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 98: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 99:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 99: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.01]
Epoch: 100:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 100: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 101:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 101: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 102:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 102: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 103:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 103: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 104:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 104: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 105:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 105: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 106:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 106: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 107:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 107: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 108:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 108: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 109:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 109: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 110:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 110: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 111:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 111: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 112:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 112: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 113:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 113: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 114:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 114: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 115:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 115: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 116:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 116: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 117:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 117: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 118:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 118: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 119:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 119: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 120:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 120: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 121:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 121: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 122:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 122: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 123:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 123: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 124:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 124: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 125:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 125: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 126:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 126: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 127:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 127: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 128:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 128: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 129:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 129: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 130:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 130: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 131:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 131: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 132:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 132: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 133:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 133: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 134:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 134: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 135:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 135: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 136:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 136: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 137:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 137: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 138:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 138: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 139:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 139: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 140:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 140: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 141:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 141: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 142:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 142: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 143:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 143: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 144:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 144: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 145:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 145: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 146:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 146: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 147:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 147: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 148:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 148: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 149:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 149: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 150:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 150: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 151:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 151: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 152:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 152: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 153:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 153: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 154:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 154: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 155:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 155: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 156:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 156: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 157:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 157: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 158:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 158: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 159:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 159: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 160:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 160: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 161:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 161: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 162:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 162: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 163:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 163: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 164:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 164: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 165:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 165: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 166:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 166: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 167:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 167: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 168:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 168: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 169:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 169: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 170:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 170: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 171:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 171: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 172:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 172: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 173:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 173: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 174:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 174: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 175:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 175: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 176:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 176: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 177:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 177: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 178:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 178: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 179:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 179: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 180:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 180: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 181:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 181: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 182:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 182: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 183:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 183: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 184:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 184: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 185:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 185: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 186:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 186: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 187:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 187: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 188:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 188: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 189:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 189: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 190:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 190: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 191:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 191: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 192:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 192: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 193:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 193: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 194:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 194: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 195:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 195: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 196:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 196: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 197:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 197: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 198:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 198: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 199:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 199: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 200:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 200: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 201:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 201: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 202:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 202: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 203:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 203: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 204:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 204: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 205:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 205: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 206:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 206: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 207:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 207: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 208:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 208: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 209:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 209: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 210:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 210: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 211:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 211: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 212:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 212: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 213:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 213: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 214:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 214: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 215:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 215: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 216:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 216: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 217:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 217: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 218:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 218: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 219:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 219: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 220:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 220: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 221:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 221: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 222:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 222: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 223:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 223: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 224:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 224: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 225:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 225: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 226:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 226: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 227:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 227: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 228:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 228: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 229:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 229: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 230:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 230: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 231:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 231: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 232:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 232: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 233:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 233: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 234:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 234: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 235:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 235: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 236:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 236: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 237:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 237: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 238:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 238: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 239:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 239: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 240:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 240: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 241:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 241: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 242:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 242: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 243:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 243: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 244:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 244: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 245:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 245: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 246:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 246: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 247:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 247: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 248:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 248: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 249:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 249: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 250:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 250: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 251:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 251: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 252:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 252: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 253:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 253: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 254:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 254: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 255:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 255: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 256:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 256: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 257:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 257: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 258:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 258: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 259:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 259: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 260:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 260: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 261:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 261: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 262:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 262: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 263:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 263: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 264:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 264: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 265:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 265: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 266:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 266: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 267:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 267: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 268:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 268: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 269:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 269: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 270:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 270: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 271:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 271: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 272:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 272: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 273:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 273: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 274:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 274: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 275:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 275: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 276:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 276: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 277:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 277: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 278:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 278: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 279:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 279: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 280:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 280: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 281:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 281: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 282:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 282: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 283:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 283: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 284:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 284: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 285:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 285: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 286:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 286: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 287:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 287: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 288:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 288: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 289:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 289: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 290:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 290: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 291:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 291: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 292:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 292: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 293:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 293: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 294:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 294: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 295:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 295: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 296:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 296: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 297:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 297: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 298:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 298: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 299:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 299: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 100% 300/300 [52:55<00:00, 10.59s/it]
INFO:root:Training is finished ...
Iteration: 100% 24/24 [00:10<00:00,  2.24it/s, acc=1]
Iteration: 100% 172/172 [20:25<00:00,  7.12s/it, acc=0.86]
obj_pp_to_subj_pp: 19.7
cp_recursion: 53.6
pp_recursion: 42.4
subj_to_obj_proper: 90.5
prim_to_obj_proper: 86.1
prim_to_subj_proper: 100.0
v_dat_p2_pp_moved_to_recipient: 4.878048780487805
LEX: 99.63333333333333
OVERALL: 86.04439592430859
INFO:root:Baselining the Transformer Encoder-Decoder Model
INFO:root:__Number CUDA Devices: 1
INFO:root:Number of model params: 4344077
INFO:root:OUTPUT DIR: ./results_cogs_train_default_wu_data_gen_on_modified_v_dat_p2/cogs_pipeline.model.ende_transformer.lf.cogs.glove.False.seed.77
Epoch: 0:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 0: 100% 213/213 [00:10<00:00, 20.10it/s, loss=6.02]
Epoch: 1:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 1: 100% 213/213 [00:10<00:00, 20.24it/s, loss=4.68]
Epoch: 2:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 2: 100% 213/213 [00:10<00:00, 20.21it/s, loss=3.62]
Epoch: 3:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 3: 100% 213/213 [00:10<00:00, 20.23it/s, loss=2.5]
Epoch: 4:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 4: 100% 213/213 [00:10<00:00, 20.16it/s, loss=1.96]
Epoch: 5:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 5: 100% 213/213 [00:10<00:00, 20.17it/s, loss=1.62]
Epoch: 6:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 6: 100% 213/213 [00:10<00:00, 20.26it/s, loss=1.36]
Epoch: 7:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 7: 100% 213/213 [00:10<00:00, 20.24it/s, loss=1.18]
Epoch: 8:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 8: 100% 213/213 [00:10<00:00, 20.26it/s, loss=1.06]
Epoch: 9:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 9: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.95]
Epoch: 10:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 10: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.86]
Epoch: 11:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 11: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0.77]
Epoch: 12:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 12: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.69]
Epoch: 13:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 13: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.62]
Epoch: 14:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 14: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.55]
Epoch: 15:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 15: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.49]
Epoch: 16:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 16: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.43]
Epoch: 17:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 17: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.37]
Epoch: 18:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 18: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.32]
Epoch: 19:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 19: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.29]
Epoch: 20:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 20: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.25]
Epoch: 21:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 21: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.22]
Epoch: 22:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 22: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.2]
Epoch: 23:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 23: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0.18]
Epoch: 24:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 24: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.16]
Epoch: 25:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 25: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.14]
Epoch: 26:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 26: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.13]
Epoch: 27:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 27: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.11]
Epoch: 28:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 28: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.11]
Epoch: 29:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 29: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0.1]
Epoch: 30:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 30: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.1]
Epoch: 31:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 31: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.08]
Epoch: 32:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 32: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.07]
Epoch: 33:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 33: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.07]
Epoch: 34:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 34: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0.07]
Epoch: 35:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 35: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.06]
Epoch: 36:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 36: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.06]
Epoch: 37:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 37: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.05]
Epoch: 38:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 38: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.05]
Epoch: 39:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 39: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.04]
Epoch: 40:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 40: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0.05]
Epoch: 41:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 41: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.04]
Epoch: 42:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 42: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.03]
Epoch: 43:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 43: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.04]
Epoch: 44:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 44: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.03]
Epoch: 45:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 45: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.03]
Epoch: 46:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 46: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0.03]
Epoch: 47:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 47: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.03]
Epoch: 48:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 48: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.02]
Epoch: 49:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 49: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.02]
Epoch: 50:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 50: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.03]
Epoch: 51:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 51: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.02]
Epoch: 52:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 52: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.03]
Epoch: 53:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 53: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.02]
Epoch: 54:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 54: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.02]
Epoch: 55:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 55: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.02]
Epoch: 56:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 56: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.02]
Epoch: 57:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 57: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.02]
Epoch: 58:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 58: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.02]
Epoch: 59:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 59: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.02]
Epoch: 60:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 60: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.02]
Epoch: 61:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 61: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.01]
Epoch: 62:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 62: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.01]
Epoch: 63:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 63: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.02]
Epoch: 64:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 64: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.02]
Epoch: 65:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 65: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.01]
Epoch: 66:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 66: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.01]
Epoch: 67:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 67: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.01]
Epoch: 68:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 68: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.01]
Epoch: 69:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 69: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.01]
Epoch: 70:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 70: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.01]
Epoch: 71:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 71: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.01]
Epoch: 72:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 72: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0.01]
Epoch: 73:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 73: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.01]
Epoch: 74:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 74: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.02]
Epoch: 75:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 75: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0.01]
Epoch: 76:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 76: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.01]
Epoch: 77:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 77: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.01]
Epoch: 78:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 78: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.01]
Epoch: 79:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 79: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.01]
Epoch: 80:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 80: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.01]
Epoch: 81:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 81: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0.01]
Epoch: 82:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 82: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.01]
Epoch: 83:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 83: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.01]
Epoch: 84:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 84: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.01]
Epoch: 85:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 85: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.01]
Epoch: 86:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 86: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.01]
Epoch: 87:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 87: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 88:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 88: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 89:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 89: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.01]
Epoch: 90:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 90: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.01]
Epoch: 91:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 91: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 92:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 92: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0.01]
Epoch: 93:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 93: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 94:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 94: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.01]
Epoch: 95:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 95: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 96:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 96: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 97:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 97: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 98:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 98: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 99:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 99: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 100:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 100: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 101:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 101: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.01]
Epoch: 102:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 102: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 103:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 103: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 104:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 104: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 105:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 105: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 106:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 106: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 107:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 107: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 108:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 108: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 109:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 109: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 110:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 110: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 111:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 111: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 112:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 112: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 113:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 113: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 114:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 114: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 115:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 115: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 116:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 116: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 117:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 117: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 118:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 118: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 119:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 119: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 120:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 120: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 121:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 121: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 122:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 122: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 123:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 123: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 124:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 124: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 125:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 125: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 126:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 126: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 127:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 127: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 128:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 128: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.01]
Epoch: 129:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 129: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 130:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 130: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 131:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 131: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 132:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 132: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 133:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 133: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 134:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 134: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 135:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 135: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 136:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 136: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 137:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 137: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 138:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 138: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 139:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 139: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.01]
Epoch: 140:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 140: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 141:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 141: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 142:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 142: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 143:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 143: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 144:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 144: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 145:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 145: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 146:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 146: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 147:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 147: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 148:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 148: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 149:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 149: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 150:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 150: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 151:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 151: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 152:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 152: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 153:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 153: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 154:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 154: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 155:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 155: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 156:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 156: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 157:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 157: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 158:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 158: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 159:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 159: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.01]
Epoch: 160:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 160: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 161:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 161: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 162:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 162: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 163:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 163: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 164:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 164: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 165:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 165: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 166:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 166: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 167:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 167: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 168:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 168: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 169:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 169: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 170:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 170: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 171:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 171: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 172:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 172: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 173:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 173: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 174:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 174: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 175:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 175: 100% 213/213 [00:10<00:00, 20.04it/s, loss=0]
Epoch: 176:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 176: 100% 213/213 [00:10<00:00, 20.06it/s, loss=0]
Epoch: 177:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 177: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 178:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 178: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 179:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 179: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 180:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 180: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 181:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 181: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 182:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 182: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 183:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 183: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 184:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 184: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 185:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 185: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 186:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 186: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 187:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 187: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 188:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 188: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 189:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 189: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 190:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 190: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 191:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 191: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 192:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 192: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 193:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 193: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 194:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 194: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 195:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 195: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 196:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 196: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 197:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 197: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 198:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 198: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 199:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 199: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 200:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 200: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 201:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 201: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 202:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 202: 100% 213/213 [00:10<00:00, 20.09it/s, loss=0]
Epoch: 203:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 203: 100% 213/213 [00:10<00:00, 19.85it/s, loss=0]
Epoch: 204:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 204: 100% 213/213 [00:10<00:00, 19.98it/s, loss=0]
Epoch: 205:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 205: 100% 213/213 [00:10<00:00, 20.03it/s, loss=0]
Epoch: 206:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 206: 100% 213/213 [00:10<00:00, 19.96it/s, loss=0]
Epoch: 207:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 207: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 208:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 208: 100% 213/213 [00:10<00:00, 20.05it/s, loss=0]
Epoch: 209:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 209: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 210:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 210: 100% 213/213 [00:10<00:00, 19.88it/s, loss=0]
Epoch: 211:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 211: 100% 213/213 [00:10<00:00, 20.08it/s, loss=0]
Epoch: 212:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 212: 100% 213/213 [00:10<00:00, 19.69it/s, loss=0]
Epoch: 213:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 213: 100% 213/213 [00:10<00:00, 19.67it/s, loss=0]
Epoch: 214:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 214: 100% 213/213 [00:10<00:00, 19.62it/s, loss=0]
Epoch: 215:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 215: 100% 213/213 [00:10<00:00, 19.70it/s, loss=0]
Epoch: 216:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 216: 100% 213/213 [00:10<00:00, 19.71it/s, loss=0]
Epoch: 217:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 217: 100% 213/213 [00:10<00:00, 19.70it/s, loss=0]
Epoch: 218:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 218: 100% 213/213 [00:10<00:00, 19.69it/s, loss=0]
Epoch: 219:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 219: 100% 213/213 [00:10<00:00, 19.71it/s, loss=0]
Epoch: 220:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 220: 100% 213/213 [00:10<00:00, 19.64it/s, loss=0]
Epoch: 221:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 221: 100% 213/213 [00:10<00:00, 19.73it/s, loss=0]
Epoch: 222:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 222: 100% 213/213 [00:10<00:00, 20.01it/s, loss=0]
Epoch: 223:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 223: 100% 213/213 [00:10<00:00, 19.96it/s, loss=0]
Epoch: 224:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 224: 100% 213/213 [00:10<00:00, 19.78it/s, loss=0]
Epoch: 225:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 225: 100% 213/213 [00:10<00:00, 19.62it/s, loss=0]
Epoch: 226:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 226: 100% 213/213 [00:10<00:00, 19.70it/s, loss=0]
Epoch: 227:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 227: 100% 213/213 [00:10<00:00, 19.70it/s, loss=0]
Epoch: 228:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 228: 100% 213/213 [00:10<00:00, 19.70it/s, loss=0]
Epoch: 229:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 229: 100% 213/213 [00:10<00:00, 19.68it/s, loss=0]
Epoch: 230:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 230: 100% 213/213 [00:10<00:00, 19.71it/s, loss=0]
Epoch: 231:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 231: 100% 213/213 [00:10<00:00, 19.66it/s, loss=0]
Epoch: 232:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 232: 100% 213/213 [00:10<00:00, 19.68it/s, loss=0]
Epoch: 233:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 233: 100% 213/213 [00:10<00:00, 19.69it/s, loss=0]
Epoch: 234:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 234: 100% 213/213 [00:10<00:00, 19.80it/s, loss=0]
Epoch: 235:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 235: 100% 213/213 [00:10<00:00, 19.96it/s, loss=0]
Epoch: 236:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 236: 100% 213/213 [00:10<00:00, 19.88it/s, loss=0]
Epoch: 237:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 237: 100% 213/213 [00:10<00:00, 19.63it/s, loss=0]
Epoch: 238:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 238: 100% 213/213 [00:10<00:00, 19.69it/s, loss=0]
Epoch: 239:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 239: 100% 213/213 [00:10<00:00, 19.71it/s, loss=0]
Epoch: 240:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 240: 100% 213/213 [00:10<00:00, 19.69it/s, loss=0]
Epoch: 241:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 241: 100% 213/213 [00:10<00:00, 19.66it/s, loss=0]
Epoch: 242:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 242: 100% 213/213 [00:10<00:00, 19.65it/s, loss=0]
Epoch: 243:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 243: 100% 213/213 [00:10<00:00, 19.68it/s, loss=0]
Epoch: 244:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 244: 100% 213/213 [00:10<00:00, 19.71it/s, loss=0]
Epoch: 245:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 245: 100% 213/213 [00:10<00:00, 19.71it/s, loss=0]
Epoch: 246:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 246: 100% 213/213 [00:10<00:00, 19.93it/s, loss=0]
Epoch: 247:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 247: 100% 213/213 [00:10<00:00, 19.95it/s, loss=0]
Epoch: 248:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 248: 100% 213/213 [00:10<00:00, 19.70it/s, loss=0]
Epoch: 249:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 249: 100% 213/213 [00:10<00:00, 19.71it/s, loss=0]
Epoch: 250:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 250: 100% 213/213 [00:10<00:00, 19.70it/s, loss=0]
Epoch: 251:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 251: 100% 213/213 [00:10<00:00, 19.68it/s, loss=0]
Epoch: 252:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 252: 100% 213/213 [00:10<00:00, 19.68it/s, loss=0]
Epoch: 253:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 253: 100% 213/213 [00:10<00:00, 19.71it/s, loss=0]
Epoch: 254:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 254: 100% 213/213 [00:10<00:00, 19.63it/s, loss=0]
Epoch: 255:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 255: 100% 213/213 [00:10<00:00, 19.67it/s, loss=0]
Epoch: 256:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 256: 100% 213/213 [00:10<00:00, 19.68it/s, loss=0]
Epoch: 257:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 257: 100% 213/213 [00:10<00:00, 19.72it/s, loss=0]
Epoch: 258:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 258: 100% 213/213 [00:10<00:00, 19.99it/s, loss=0]
Epoch: 259:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 259: 100% 213/213 [00:10<00:00, 19.86it/s, loss=0]
Epoch: 260:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 260: 100% 213/213 [00:10<00:00, 19.73it/s, loss=0]
Epoch: 261:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 261: 100% 213/213 [00:10<00:00, 19.69it/s, loss=0]
Epoch: 262:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 262: 100% 213/213 [00:10<00:00, 19.67it/s, loss=0]
Epoch: 263:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 263: 100% 213/213 [00:10<00:00, 19.68it/s, loss=0]
Epoch: 264:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 264: 100% 213/213 [00:10<00:00, 19.65it/s, loss=0]
Epoch: 265:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 265: 100% 213/213 [00:10<00:00, 19.56it/s, loss=0]
Epoch: 266:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 266: 100% 213/213 [00:10<00:00, 19.68it/s, loss=0]
Epoch: 267:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 267: 100% 213/213 [00:10<00:00, 19.69it/s, loss=0]
Epoch: 268:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 268: 100% 213/213 [00:10<00:00, 19.68it/s, loss=0]
Epoch: 269:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 269: 100% 213/213 [00:10<00:00, 19.85it/s, loss=0]
Epoch: 270:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 270: 100% 213/213 [00:10<00:00, 19.97it/s, loss=0]
Epoch: 271:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 271: 100% 213/213 [00:10<00:00, 19.76it/s, loss=0]
Epoch: 272:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 272: 100% 213/213 [00:10<00:00, 19.68it/s, loss=0]
Epoch: 273:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 273: 100% 213/213 [00:10<00:00, 19.67it/s, loss=0]
Epoch: 274:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 274: 100% 213/213 [00:10<00:00, 19.70it/s, loss=0]
Epoch: 275:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 275: 100% 213/213 [00:10<00:00, 19.70it/s, loss=0]
Epoch: 276:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 276: 100% 213/213 [00:10<00:00, 19.64it/s, loss=0]
Epoch: 277:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 277: 100% 213/213 [00:10<00:00, 19.70it/s, loss=0]
Epoch: 278:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 278: 100% 213/213 [00:10<00:00, 19.69it/s, loss=0]
Epoch: 279:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 279: 100% 213/213 [00:10<00:00, 19.69it/s, loss=0]
Epoch: 280:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 280: 100% 213/213 [00:10<00:00, 19.69it/s, loss=0]
Epoch: 281:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 281: 100% 213/213 [00:10<00:00, 19.98it/s, loss=0]
Epoch: 282:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 282: 100% 213/213 [00:10<00:00, 19.89it/s, loss=0]
Epoch: 283:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 283: 100% 213/213 [00:10<00:00, 19.72it/s, loss=0]
Epoch: 284:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 284: 100% 213/213 [00:10<00:00, 19.70it/s, loss=0]
Epoch: 285:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 285: 100% 213/213 [00:10<00:00, 19.70it/s, loss=0]
Epoch: 286:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 286: 100% 213/213 [00:10<00:00, 19.66it/s, loss=0]
Epoch: 287:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 287: 100% 213/213 [00:10<00:00, 19.69it/s, loss=0]
Epoch: 288:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 288: 100% 213/213 [00:10<00:00, 19.64it/s, loss=0]
Epoch: 289:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 289: 100% 213/213 [00:10<00:00, 19.69it/s, loss=0]
Epoch: 290:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 290: 100% 213/213 [00:10<00:00, 19.71it/s, loss=0]
Epoch: 291:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 291: 100% 213/213 [00:10<00:00, 19.68it/s, loss=0]
Epoch: 292:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 292: 100% 213/213 [00:10<00:00, 19.83it/s, loss=0]
Epoch: 293:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 293: 100% 213/213 [00:10<00:00, 19.94it/s, loss=0]
Epoch: 294:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 294: 100% 213/213 [00:10<00:00, 19.85it/s, loss=0]
Epoch: 295:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 295: 100% 213/213 [00:10<00:00, 19.68it/s, loss=0]
Epoch: 296:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 296: 100% 213/213 [00:10<00:00, 19.71it/s, loss=0]
Epoch: 297:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 297: 100% 213/213 [00:10<00:00, 19.72it/s, loss=0]
Epoch: 298:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 298: 100% 213/213 [00:10<00:00, 19.71it/s, loss=0]
Epoch: 299:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 299: 100% 213/213 [00:10<00:00, 19.65it/s, loss=0]
Epoch: 100% 300/300 [53:23<00:00, 10.68s/it]
INFO:root:Training is finished ...
Iteration: 100% 24/24 [00:11<00:00,  2.10it/s, acc=1]
Iteration: 100% 172/172 [17:57<00:00,  6.26s/it, acc=0.866]
obj_pp_to_subj_pp: 31.1
cp_recursion: 53.7
pp_recursion: 43.1
subj_to_obj_proper: 88.0
prim_to_obj_proper: 92.4
prim_to_subj_proper: 99.9
v_dat_p2_pp_moved_to_recipient: 11.585365853658537
LEX: 98.99333333333334
OVERALL: 86.63118631732168
INFO:root:Baselining the Transformer Encoder-Decoder Model
INFO:root:__Number CUDA Devices: 1
INFO:root:Number of model params: 4344077
INFO:root:OUTPUT DIR: ./results_cogs_train_default_wu_data_gen_on_modified_v_dat_p2/cogs_pipeline.model.ende_transformer.lf.cogs.glove.False.seed.88
Epoch: 0:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 0: 100% 213/213 [00:10<00:00, 19.65it/s, loss=5.98]
Epoch: 1:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 1: 100% 213/213 [00:10<00:00, 19.88it/s, loss=4.6]
Epoch: 2:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 2: 100% 213/213 [00:10<00:00, 19.89it/s, loss=3.51]
Epoch: 3:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 3: 100% 213/213 [00:10<00:00, 19.90it/s, loss=2.38]
Epoch: 4:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 4: 100% 213/213 [00:10<00:00, 19.92it/s, loss=1.81]
Epoch: 5:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 5: 100% 213/213 [00:10<00:00, 19.96it/s, loss=1.48]
Epoch: 6:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 6: 100% 213/213 [00:10<00:00, 20.00it/s, loss=1.2]
Epoch: 7:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 7: 100% 213/213 [00:10<00:00, 20.06it/s, loss=0.99]
Epoch: 8:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 8: 100% 213/213 [00:10<00:00, 19.89it/s, loss=0.86]
Epoch: 9:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 9: 100% 213/213 [00:10<00:00, 19.87it/s, loss=0.77]
Epoch: 10:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 10: 100% 213/213 [00:10<00:00, 19.90it/s, loss=0.68]
Epoch: 11:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 11: 100% 213/213 [00:10<00:00, 19.87it/s, loss=0.58]
Epoch: 12:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 12: 100% 213/213 [00:10<00:00, 19.82it/s, loss=0.51]
Epoch: 13:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 13: 100% 213/213 [00:10<00:00, 19.88it/s, loss=0.43]
Epoch: 14:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 14: 100% 213/213 [00:10<00:00, 19.89it/s, loss=0.36]
Epoch: 15:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 15: 100% 213/213 [00:10<00:00, 19.88it/s, loss=0.3]
Epoch: 16:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 16: 100% 213/213 [00:10<00:00, 19.91it/s, loss=0.25]
Epoch: 17:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 17: 100% 213/213 [00:10<00:00, 19.94it/s, loss=0.19]
Epoch: 18:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 18: 100% 213/213 [00:10<00:00, 20.08it/s, loss=0.16]
Epoch: 19:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 19: 100% 213/213 [00:10<00:00, 19.89it/s, loss=0.12]
Epoch: 20:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 20: 100% 213/213 [00:10<00:00, 19.85it/s, loss=0.1]
Epoch: 21:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 21: 100% 213/213 [00:10<00:00, 19.86it/s, loss=0.08]
Epoch: 22:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 22: 100% 213/213 [00:10<00:00, 19.90it/s, loss=0.07]
Epoch: 23:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 23: 100% 213/213 [00:10<00:00, 19.83it/s, loss=0.06]
Epoch: 24:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 24: 100% 213/213 [00:10<00:00, 19.90it/s, loss=0.04]
Epoch: 25:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 25: 100% 213/213 [00:10<00:00, 19.88it/s, loss=0.04]
Epoch: 26:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 26: 100% 213/213 [00:10<00:00, 19.86it/s, loss=0.03]
Epoch: 27:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 27: 100% 213/213 [00:10<00:00, 19.89it/s, loss=0.03]
Epoch: 28:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 28: 100% 213/213 [00:10<00:00, 19.97it/s, loss=0.03]
Epoch: 29:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 29: 100% 213/213 [00:10<00:00, 19.97it/s, loss=0.02]
Epoch: 30:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 30: 100% 213/213 [00:10<00:00, 19.97it/s, loss=0.02]
Epoch: 31:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 31: 100% 213/213 [00:10<00:00, 19.84it/s, loss=0.01]
Epoch: 32:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 32: 100% 213/213 [00:10<00:00, 19.89it/s, loss=0.01]
Epoch: 33:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 33: 100% 213/213 [00:10<00:00, 19.90it/s, loss=0.01]
Epoch: 34:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 34: 100% 213/213 [00:10<00:00, 19.90it/s, loss=0.01]
Epoch: 35:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 35: 100% 213/213 [00:10<00:00, 19.83it/s, loss=0.01]
Epoch: 36:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 36: 100% 213/213 [00:10<00:00, 19.91it/s, loss=0.01]
Epoch: 37:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 37: 100% 213/213 [00:10<00:00, 19.87it/s, loss=0.01]
Epoch: 38:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 38: 100% 213/213 [00:10<00:00, 19.93it/s, loss=0.01]
Epoch: 39:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 39: 100% 213/213 [00:10<00:00, 19.96it/s, loss=0.01]
Epoch: 40:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 40: 100% 213/213 [00:10<00:00, 20.00it/s, loss=0]
Epoch: 41:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 41: 100% 213/213 [00:10<00:00, 20.06it/s, loss=0]
Epoch: 42:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 42: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0.01]
Epoch: 43:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 43: 100% 213/213 [00:10<00:00, 20.10it/s, loss=0.01]
Epoch: 44:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 44: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 45:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 45: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 46:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 46: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.01]
Epoch: 47:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 47: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 48:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 48: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.01]
Epoch: 49:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 49: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 50:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 50: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 51:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 51: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 52:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 52: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 53:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 53: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 54:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 54: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 55:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 55: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 56:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 56: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 57:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 57: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 58:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 58: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 59:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 59: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 60:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 60: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 61:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 61: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 62:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 62: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 63:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 63: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 64:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 64: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 65:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 65: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 66:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 66: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 67:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 67: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 68:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 68: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 69:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 69: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 70:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 70: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 71:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 71: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 72:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 72: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 73:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 73: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 74:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 74: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 75:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 75: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 76:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 76: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 77:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 77: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 78:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 78: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 79:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 79: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 80:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 80: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 81:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 81: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 82:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 82: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 83:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 83: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 84:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 84: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 85:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 85: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 86:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 86: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 87:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 87: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 88:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 88: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 89:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 89: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 90:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 90: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 91:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 91: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 92:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 92: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 93:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 93: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 94:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 94: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 95:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 95: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 96:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 96: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 97:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 97: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 98:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 98: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 99:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 99: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 100:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 100: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 101:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 101: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 102:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 102: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 103:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 103: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 104:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 104: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 105:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 105: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 106:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 106: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 107:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 107: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 108:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 108: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 109:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 109: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 110:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 110: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 111:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 111: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 112:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 112: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 113:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 113: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 114:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 114: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 115:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 115: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 116:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 116: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 117:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 117: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 118:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 118: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 119:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 119: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 120:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 120: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 121:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 121: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 122:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 122: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 123:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 123: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 124:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 124: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 125:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 125: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 126:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 126: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 127:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 127: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 128:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 128: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 129:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 129: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 130:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 130: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 131:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 131: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 132:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 132: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 133:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 133: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 134:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 134: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 135:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 135: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 136:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 136: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 137:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 137: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 138:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 138: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 139:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 139: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 140:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 140: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 141:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 141: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 142:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 142: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 143:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 143: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 144:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 144: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 145:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 145: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 146:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 146: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 147:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 147: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 148:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 148: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 149:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 149: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 150:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 150: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 151:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 151: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 152:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 152: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 153:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 153: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 154:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 154: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 155:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 155: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 156:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 156: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 157:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 157: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 158:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 158: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 159:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 159: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 160:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 160: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 161:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 161: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 162:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 162: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 163:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 163: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 164:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 164: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 165:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 165: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 166:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 166: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 167:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 167: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 168:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 168: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 169:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 169: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 170:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 170: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 171:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 171: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 172:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 172: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 173:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 173: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 174:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 174: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 175:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 175: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 176:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 176: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 177:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 177: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 178:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 178: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 179:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 179: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 180:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 180: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 181:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 181: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 182:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 182: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 183:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 183: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 184:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 184: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 185:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 185: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 186:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 186: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 187:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 187: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 188:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 188: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 189:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 189: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 190:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 190: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 191:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 191: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 192:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 192: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 193:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 193: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 194:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 194: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 195:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 195: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 196:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 196: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 197:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 197: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 198:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 198: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 199:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 199: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 200:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 200: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 201:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 201: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 202:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 202: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 203:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 203: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 204:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 204: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 205:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 205: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 206:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 206: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 207:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 207: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 208:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 208: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 209:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 209: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 210:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 210: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 211:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 211: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 212:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 212: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 213:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 213: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 214:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 214: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 215:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 215: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 216:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 216: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 217:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 217: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 218:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 218: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 219:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 219: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 220:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 220: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 221:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 221: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 222:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 222: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 223:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 223: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 224:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 224: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 225:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 225: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 226:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 226: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 227:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 227: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 228:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 228: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 229:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 229: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 230:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 230: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 231:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 231: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 232:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 232: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 233:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 233: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 234:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 234: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 235:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 235: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 236:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 236: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 237:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 237: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 238:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 238: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 239:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 239: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 240:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 240: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 241:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 241: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 242:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 242: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 243:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 243: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 244:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 244: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 245:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 245: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 246:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 246: 100% 213/213 [00:10<00:00, 20.10it/s, loss=0]
Epoch: 247:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 247: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 248:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 248: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 249:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 249: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 250:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 250: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 251:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 251: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 252:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 252: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 253:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 253: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 254:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 254: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 255:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 255: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 256:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 256: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 257:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 257: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 258:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 258: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 259:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 259: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 260:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 260: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 261:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 261: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 262:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 262: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 263:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 263: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 264:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 264: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 265:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 265: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 266:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 266: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 267:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 267: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 268:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 268: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 269:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 269: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 270:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 270: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 271:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 271: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 272:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 272: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 273:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 273: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 274:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 274: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 275:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 275: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 276:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 276: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 277:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 277: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 278:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 278: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 279:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 279: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 280:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 280: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 281:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 281: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 282:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 282: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 283:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 283: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 284:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 284: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 285:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 285: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 286:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 286: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 287:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 287: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 288:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 288: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 289:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 289: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 290:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 290: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 291:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 291: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 292:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 292: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 293:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 293: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 294:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 294: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 295:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 295: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 296:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 296: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 297:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 297: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 298:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 298: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 299:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 299: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 100% 300/300 [52:48<00:00, 10.56s/it]
INFO:root:Training is finished ...
Iteration: 100% 24/24 [00:10<00:00,  2.24it/s, acc=1]
Iteration: 100% 172/172 [22:23<00:00,  7.81s/it, acc=0.817]
obj_pp_to_subj_pp: 13.6
cp_recursion: 50.0
pp_recursion: 21.4
subj_to_obj_proper: 90.6
prim_to_obj_proper: 90.1
prim_to_subj_proper: 100.0
v_dat_p2_pp_moved_to_recipient: 9.451219512195122
LEX: 94.69333333333333
OVERALL: 81.66848617176127
INFO:root:Baselining the Transformer Encoder-Decoder Model
INFO:root:__Number CUDA Devices: 1
INFO:root:Number of model params: 4344077
INFO:root:OUTPUT DIR: ./results_cogs_train_default_wu_data_gen_on_modified_v_dat_p2/cogs_pipeline.model.ende_transformer.lf.cogs.glove.False.seed.99
Epoch: 0:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 0: 100% 213/213 [00:10<00:00, 20.10it/s, loss=5.87]
Epoch: 1:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 1: 100% 213/213 [00:10<00:00, 20.21it/s, loss=4.55]
Epoch: 2:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 2: 100% 213/213 [00:10<00:00, 20.12it/s, loss=3.48]
Epoch: 3:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 3: 100% 213/213 [00:10<00:00, 20.16it/s, loss=2.44]
Epoch: 4:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 4: 100% 213/213 [00:10<00:00, 20.18it/s, loss=1.91]
Epoch: 5:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 5: 100% 213/213 [00:10<00:00, 20.21it/s, loss=1.58]
Epoch: 6:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 6: 100% 213/213 [00:10<00:00, 20.20it/s, loss=1.32]
Epoch: 7:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 7: 100% 213/213 [00:10<00:00, 20.23it/s, loss=1.14]
Epoch: 8:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 8: 100% 213/213 [00:10<00:00, 20.13it/s, loss=1.01]
Epoch: 9:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 9: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0.89]
Epoch: 10:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 10: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0.79]
Epoch: 11:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 11: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.7]
Epoch: 12:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 12: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.61]
Epoch: 13:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 13: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.53]
Epoch: 14:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 14: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0.46]
Epoch: 15:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 15: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.39]
Epoch: 16:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 16: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.33]
Epoch: 17:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 17: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.28]
Epoch: 18:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 18: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.23]
Epoch: 19:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 19: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.2]
Epoch: 20:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 20: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0.16]
Epoch: 21:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 21: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0.14]
Epoch: 22:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 22: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.11]
Epoch: 23:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 23: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.09]
Epoch: 24:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 24: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.08]
Epoch: 25:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 25: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.07]
Epoch: 26:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 26: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0.06]
Epoch: 27:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 27: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.05]
Epoch: 28:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 28: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0.04]
Epoch: 29:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 29: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.04]
Epoch: 30:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 30: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.04]
Epoch: 31:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 31: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0.03]
Epoch: 32:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 32: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.03]
Epoch: 33:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 33: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.03]
Epoch: 34:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 34: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.02]
Epoch: 35:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 35: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.02]
Epoch: 36:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 36: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.02]
Epoch: 37:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 37: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0.02]
Epoch: 38:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 38: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.01]
Epoch: 39:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 39: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.01]
Epoch: 40:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 40: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.01]
Epoch: 41:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 41: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.01]
Epoch: 42:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 42: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0.01]
Epoch: 43:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 43: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0.01]
Epoch: 44:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 44: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.01]
Epoch: 45:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 45: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0.01]
Epoch: 46:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 46: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0.01]
Epoch: 47:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 47: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0.01]
Epoch: 48:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 48: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.01]
Epoch: 49:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 49: 100% 213/213 [00:10<00:00, 20.08it/s, loss=0.01]
Epoch: 50:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 50: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0.01]
Epoch: 51:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 51: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0.01]
Epoch: 52:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 52: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0.01]
Epoch: 53:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 53: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 54:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 54: 100% 213/213 [00:10<00:00, 19.90it/s, loss=0]
Epoch: 55:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 55: 100% 213/213 [00:10<00:00, 20.05it/s, loss=0.01]
Epoch: 56:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 56: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.01]
Epoch: 57:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 57: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 58:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 58: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 59:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 59: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0.01]
Epoch: 60:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 60: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0.01]
Epoch: 61:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 61: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 62:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 62: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 63:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 63: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 64:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 64: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 65:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 65: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 66:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 66: 100% 213/213 [00:10<00:00, 20.10it/s, loss=0]
Epoch: 67:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 67: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 68:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 68: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 69:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 69: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 70:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 70: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 71:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 71: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 72:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 72: 100% 213/213 [00:10<00:00, 20.06it/s, loss=0]
Epoch: 73:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 73: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 74:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 74: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 75:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 75: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 76:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 76: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0.01]
Epoch: 77:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 77: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 78:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 78: 100% 213/213 [00:10<00:00, 20.05it/s, loss=0]
Epoch: 79:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 79: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 80:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 80: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 81:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 81: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 82:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 82: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 83:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 83: 100% 213/213 [00:10<00:00, 20.08it/s, loss=0]
Epoch: 84:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 84: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 85:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 85: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 86:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 86: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 87:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 87: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 88:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 88: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 89:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 89: 100% 213/213 [00:10<00:00, 20.09it/s, loss=0]
Epoch: 90:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 90: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 91:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 91: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 92:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 92: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 93:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 93: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 94:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 94: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 95:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 95: 100% 213/213 [00:10<00:00, 20.10it/s, loss=0]
Epoch: 96:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 96: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 97:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 97: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 98:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 98: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 99:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 99: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 100:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 100: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 101:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 101: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 102:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 102: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 103:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 103: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 104:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 104: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 105:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 105: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 106:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 106: 100% 213/213 [00:10<00:00, 20.09it/s, loss=0]
Epoch: 107:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 107: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 108:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 108: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 109:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 109: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 110:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 110: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 111:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 111: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 112:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 112: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 113:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 113: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 114:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 114: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 115:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 115: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 116:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 116: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 117:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 117: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 118:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 118: 100% 213/213 [00:10<00:00, 20.10it/s, loss=0]
Epoch: 119:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 119: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 120:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 120: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 121:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 121: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 122:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 122: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 123:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 123: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 124:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 124: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 125:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 125: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 126:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 126: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 127:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 127: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 128:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 128: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 129:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 129: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 130:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 130: 100% 213/213 [00:10<00:00, 20.09it/s, loss=0]
Epoch: 131:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 131: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 132:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 132: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 133:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 133: 100% 213/213 [00:10<00:00, 20.06it/s, loss=0]
Epoch: 134:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 134: 100% 213/213 [00:10<00:00, 19.96it/s, loss=0]
Epoch: 135:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 135: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 136:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 136: 100% 213/213 [00:10<00:00, 20.08it/s, loss=0]
Epoch: 137:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 137: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 138:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 138: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 139:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 139: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 140:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 140: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 141:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 141: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 142:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 142: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 143:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 143: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 144:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 144: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 145:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 145: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 146:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 146: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 147:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 147: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 148:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 148: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 149:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 149: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 150:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 150: 100% 213/213 [00:10<00:00, 20.09it/s, loss=0]
Epoch: 151:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 151: 100% 213/213 [00:10<00:00, 20.04it/s, loss=0]
Epoch: 152:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 152: 100% 213/213 [00:10<00:00, 20.07it/s, loss=0]
Epoch: 153:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 153: 100% 213/213 [00:10<00:00, 20.07it/s, loss=0]
Epoch: 154:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 154: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 155:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 155: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 156:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 156: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 157:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 157: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 158:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 158: 100% 213/213 [00:10<00:00, 20.07it/s, loss=0]
Epoch: 159:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 159: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 160:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 160: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 161:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 161: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 162:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 162: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 163:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 163: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 164:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 164: 100% 213/213 [00:10<00:00, 20.10it/s, loss=0]
Epoch: 165:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 165: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 166:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 166: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 167:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 167: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 168:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 168: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 169:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 169: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 170:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 170: 100% 213/213 [00:10<00:00, 20.09it/s, loss=0]
Epoch: 171:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 171: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 172:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 172: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 173:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 173: 100% 213/213 [00:10<00:00, 20.08it/s, loss=0]
Epoch: 174:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 174: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 175:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 175: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 176:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 176: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 177:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 177: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 178:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 178: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 179:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 179: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 180:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 180: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 181:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 181: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 182:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 182: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 183:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 183: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 184:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 184: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 185:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 185: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 186:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 186: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 187:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 187: 100% 213/213 [00:10<00:00, 20.10it/s, loss=0]
Epoch: 188:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 188: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 189:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 189: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 190:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 190: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 191:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 191: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 192:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 192: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 193:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 193: 100% 213/213 [00:10<00:00, 20.04it/s, loss=0]
Epoch: 194:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 194: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 195:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 195: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 196:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 196: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 197:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 197: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 198:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 198: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 199:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 199: 100% 213/213 [00:10<00:00, 20.10it/s, loss=0]
Epoch: 200:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 200: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 201:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 201: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 202:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 202: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 203:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 203: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 204:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 204: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 205:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 205: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 206:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 206: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 207:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 207: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 208:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 208: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 209:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 209: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 210:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 210: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 211:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 211: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 212:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 212: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 213:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 213: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 214:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 214: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 215:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 215: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 216:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 216: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 217:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 217: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 218:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 218: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 219:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 219: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 220:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 220: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 221:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 221: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 222:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 222: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 223:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 223: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 224:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 224: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 225:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 225: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 226:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 226: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 227:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 227: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 228:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 228: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 229:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 229: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 230:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 230: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 231:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 231: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 232:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 232: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 233:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 233: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 234:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 234: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 235:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 235: 100% 213/213 [00:10<00:00, 20.10it/s, loss=0]
Epoch: 236:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 236: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 237:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 237: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 238:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 238: 100% 213/213 [00:10<00:00, 20.08it/s, loss=0]
Epoch: 239:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 239: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 240:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 240: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 241:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 241: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 242:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 242: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 243:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 243: 100% 213/213 [00:10<00:00, 20.09it/s, loss=0]
Epoch: 244:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 244: 100% 213/213 [00:10<00:00, 19.98it/s, loss=0]
Epoch: 245:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 245: 100% 213/213 [00:10<00:00, 19.89it/s, loss=0]
Epoch: 246:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 246: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 247:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 247: 100% 213/213 [00:10<00:00, 20.01it/s, loss=0]
Epoch: 248:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 248: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 249:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 249: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 250:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 250: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 251:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 251: 100% 213/213 [00:10<00:00, 20.08it/s, loss=0]
Epoch: 252:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 252: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 253:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 253: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 254:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 254: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 255:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 255: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 256:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 256: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 257:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 257: 100% 213/213 [00:10<00:00, 20.10it/s, loss=0]
Epoch: 258:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 258: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 259:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 259: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 260:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 260: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 261:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 261: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 262:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 262: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 263:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 263: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 264:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 264: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 265:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 265: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 266:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 266: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 267:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 267: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 268:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 268: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 269:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 269: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 270:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 270: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 271:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 271: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 272:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 272: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 273:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 273: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 274:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 274: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 275:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 275: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 276:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 276: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 277:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 277: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 278:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 278: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 279:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 279: 100% 213/213 [00:10<00:00, 20.10it/s, loss=0]
Epoch: 280:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 280: 100% 213/213 [00:10<00:00, 20.09it/s, loss=0]
Epoch: 281:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 281: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 282:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 282: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 283:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 283: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 284:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 284: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 285:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 285: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 286:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 286: 100% 213/213 [00:10<00:00, 20.05it/s, loss=0]
Epoch: 287:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 287: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 288:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 288: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 289:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 289: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 290:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 290: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 291:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 291: 100% 213/213 [00:10<00:00, 20.09it/s, loss=0]
Epoch: 292:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 292: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 293:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 293: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 294:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 294: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 295:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 295: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 296:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 296: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 297:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 297: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 298:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 298: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 299:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 299: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 100% 300/300 [53:09<00:00, 10.63s/it]
INFO:root:Training is finished ...
Iteration: 100% 24/24 [00:10<00:00,  2.24it/s, acc=1]
Iteration: 100% 172/172 [11:47<00:00,  4.11s/it, acc=0.82]
obj_pp_to_subj_pp: 18.3
cp_recursion: 51.3
pp_recursion: 48.2
subj_to_obj_proper: 94.2
prim_to_obj_proper: 91.1
prim_to_subj_proper: 100.0
v_dat_p2_pp_moved_to_recipient: 0.0
LEX: 93.24
OVERALL: 81.95505822416303

great this is consistent with the hypothesis that the v_dat_p2 pp on recipient generalization is difficult, similar to the obj_pp_to_subj_pp (if not trained for it)

scores_wu_et_al_2023_five_default_config_runs_generalize_to_v_dat_p2_recipient_pp_very_weak_performance_as_expected.png

possible_issue_with_subj_pp_generalization_by_transformers_could_be_simple_nontree_pp_np_distractor_when_modifying_nps_with_related_nps_to_right.png

got restarted by colab without custom edit above and without the TSV for 2nd set of 5 seeds, first run had it; now I run it again as I want 10 seeds

In [ ]:
!python run_cogs.py --model_name ende_transformer --use_iiem --gpu 1 --train_batch_size 128 --eval_batch_size 128 --lr 0.0001 --data_path ./cogs_train_default_wu_data_gen_on_modified_v_dat_p2 --output_dir ./results_cogs_train_default_wu_data_gen_on_modified_v_dat_p2 --lfs cogs --do_train --do_test --do_gen --max_seq_len 512 --output_json --epochs 300 --seeds "43;67;78;89;100" # paper's seeds + 1, not sure how they chose them
INFO:root:Baselining the Transformer Encoder-Decoder Model
INFO:root:__Number CUDA Devices: 1
INFO:root:Number of model params: 4344077
INFO:root:OUTPUT DIR: ./results_cogs_train_default_wu_data_gen_on_modified_v_dat_p2/cogs_pipeline.model.ende_transformer.lf.cogs.glove.False.seed.43
Epoch: 0:   0% 0/213 [00:00<?, ?it/s]We strongly recommend passing in an `attention_mask` since your input_ids may be padded. See https://huggingface.co/docs/transformers/troubleshooting#incorrect-output-when-padding-tokens-arent-masked.
/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 0: 100% 213/213 [00:12<00:00, 17.29it/s, loss=6]
Epoch: 1:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 1: 100% 213/213 [00:10<00:00, 20.27it/s, loss=4.59]
Epoch: 2:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 2: 100% 213/213 [00:10<00:00, 20.19it/s, loss=3.54]
Epoch: 3:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 3: 100% 213/213 [00:10<00:00, 20.25it/s, loss=2.48]
Epoch: 4:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 4: 100% 213/213 [00:10<00:00, 20.28it/s, loss=1.92]
Epoch: 5:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 5: 100% 213/213 [00:10<00:00, 20.24it/s, loss=1.55]
Epoch: 6:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 6: 100% 213/213 [00:10<00:00, 20.35it/s, loss=1.26]
Epoch: 7:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 7: 100% 213/213 [00:10<00:00, 20.33it/s, loss=1.08]
Epoch: 8:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 8: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.96]
Epoch: 9:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 9: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0.87]
Epoch: 10:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 10: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.77]
Epoch: 11:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 11: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.68]
Epoch: 12:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 12: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0.59]
Epoch: 13:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 13: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.51]
Epoch: 14:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 14: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0.45]
Epoch: 15:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 15: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.38]
Epoch: 16:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 16: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.32]
Epoch: 17:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 17: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.27]
Epoch: 18:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 18: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.22]
Epoch: 19:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 19: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.19]
Epoch: 20:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 20: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.15]
Epoch: 21:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 21: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.12]
Epoch: 22:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 22: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.1]
Epoch: 23:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 23: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.08]
Epoch: 24:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 24: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.06]
Epoch: 25:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 25: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0.06]
Epoch: 26:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 26: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0.04]
Epoch: 27:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 27: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.03]
Epoch: 28:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 28: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.03]
Epoch: 29:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 29: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.03]
Epoch: 30:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 30: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0.03]
Epoch: 31:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 31: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.03]
Epoch: 32:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 32: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0.02]
Epoch: 33:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 33: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.02]
Epoch: 34:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 34: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0.02]
Epoch: 35:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 35: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 36:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 36: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.01]
Epoch: 37:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 37: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.01]
Epoch: 38:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 38: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.01]
Epoch: 39:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 39: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.01]
Epoch: 40:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 40: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.01]
Epoch: 41:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 41: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.01]
Epoch: 42:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 42: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.01]
Epoch: 43:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 43: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.01]
Epoch: 44:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 44: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.01]
Epoch: 45:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 45: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.01]
Epoch: 46:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 46: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 47:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 47: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 48:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 48: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 49:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 49: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 50:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 50: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 51:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 51: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 52:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 52: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0.01]
Epoch: 53:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 53: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 54:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 54: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 55:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 55: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 56:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 56: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.01]
Epoch: 57:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 57: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 58:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 58: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.01]
Epoch: 59:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 59: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 60:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 60: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 61:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 61: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 62:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 62: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 63:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 63: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 64:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 64: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 65:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 65: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 66:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 66: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 67:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 67: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 68:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 68: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 69:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 69: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 70:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 70: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 71:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 71: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 72:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 72: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 73:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 73: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 74:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 74: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 75:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 75: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 76:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 76: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 77:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 77: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 78:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 78: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 79:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 79: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 80:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 80: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 81:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 81: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 82:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 82: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 83:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 83: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 84:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 84: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 85:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 85: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 86:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 86: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 87:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 87: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 88:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 88: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 89:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 89: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 90:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 90: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 91:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 91: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 92:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 92: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 93:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 93: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 94:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 94: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 95:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 95: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 96:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 96: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 97:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 97: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 98:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 98: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 99:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 99: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 100:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 100: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 101:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 101: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 102:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 102: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 103:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 103: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 104:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 104: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 105:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 105: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 106:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 106: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 107:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 107: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 108:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 108: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 109:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 109: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 110:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 110: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 111:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 111: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 112:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 112: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 113:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 113: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 114:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 114: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 115:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 115: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 116:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 116: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 117:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 117: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 118:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 118: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 119:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 119: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 120:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 120: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 121:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 121: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 122:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 122: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 123:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 123: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 124:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 124: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 125:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 125: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 126:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 126: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 127:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 127: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 128:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 128: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 129:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 129: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 130:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 130: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 131:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 131: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 132:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 132: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 133:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 133: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 134:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 134: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 135:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 135: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 136:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 136: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 137:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 137: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 138:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 138: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 139:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 139: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 140:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 140: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 141:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 141: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 142:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 142: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 143:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 143: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 144:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 144: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 145:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 145: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 146:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 146: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 147:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 147: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 148:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 148: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 149:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 149: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 150:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 150: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 151:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 151: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 152:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 152: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 153:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 153: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 154:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 154: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 155:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 155: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 156:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 156: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 157:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 157: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 158:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 158: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 159:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 159: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 160:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 160: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 161:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 161: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 162:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 162: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 163:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 163: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 164:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 164: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 165:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 165: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 166:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 166: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 167:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 167: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 168:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 168: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 169:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 169: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 170:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 170: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 171:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 171: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 172:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 172: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 173:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 173: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 174:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 174: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 175:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 175: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 176:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 176: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 177:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 177: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 178:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 178: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 179:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 179: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 180:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 180: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 181:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 181: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 182:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 182: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 183:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 183: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 184:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 184: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 185:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 185: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 186:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 186: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 187:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 187: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 188:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 188: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 189:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 189: 100% 213/213 [00:10<00:00, 20.03it/s, loss=0]
Epoch: 190:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 190: 100% 213/213 [00:10<00:00, 20.11it/s, loss=0]
Epoch: 191:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 191: 100% 213/213 [00:10<00:00, 20.08it/s, loss=0]
Epoch: 192:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 192: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 193:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 193: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 194:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 194: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 195:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 195: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 196:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 196: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 197:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 197: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 198:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 198: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 199:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 199: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 200:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 200: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 201:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 201: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 202:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 202: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 203:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 203: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 204:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 204: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 205:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 205: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 206:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 206: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 207:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 207: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 208:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 208: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 209:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 209: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 210:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 210: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 211:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 211: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 212:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 212: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 213:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 213: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 214:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 214: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 215:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 215: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 216:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 216: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 217:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 217: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 218:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 218: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 219:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 219: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 220:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 220: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 221:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 221: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 222:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 222: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 223:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 223: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 224:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 224: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 225:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 225: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 226:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 226: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 227:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 227: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 228:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 228: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 229:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 229: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 230:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 230: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 231:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 231: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 232:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 232: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 233:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 233: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 234:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 234: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 235:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 235: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 236:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 236: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 237:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 237: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 238:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 238: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 239:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 239: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 240:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 240: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 241:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 241: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 242:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 242: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 243:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 243: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 244:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 244: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 245:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 245: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 246:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 246: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 247:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 247: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 248:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 248: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 249:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 249: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 250:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 250: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 251:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 251: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 252:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 252: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 253:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 253: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 254:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 254: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 255:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 255: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 256:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 256: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 257:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 257: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 258:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 258: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 259:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 259: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 260:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 260: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 261:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 261: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 262:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 262: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 263:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 263: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 264:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 264: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 265:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 265: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 266:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 266: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 267:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 267: 100% 213/213 [00:10<00:00, 20.06it/s, loss=0]
Epoch: 268:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 268: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 269:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 269: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 270:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 270: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 271:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 271: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 272:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 272: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 273:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 273: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 274:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 274: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 275:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 275: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 276:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 276: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 277:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 277: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 278:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 278: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 279:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 279: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 280:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 280: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 281:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 281: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 282:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 282: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 283:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 283: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 284:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 284: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 285:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 285: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 286:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 286: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 287:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 287: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 288:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 288: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 289:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 289: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 290:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 290: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 291:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 291: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 292:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 292: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 293:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 293: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 294:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 294: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 295:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 295: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 296:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 296: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 297:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 297: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 298:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 298: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 299:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 299: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 100% 300/300 [52:48<00:00, 10.56s/it]
INFO:root:Training is finished ...
Iteration: 100% 24/24 [00:10<00:00,  2.26it/s, acc=1]
Iteration: 100% 167/167 [18:01<00:00,  6.47s/it, acc=0.893]
obj_pp_to_subj_pp: 20.2
cp_recursion: 52.0
pp_recursion: 61.8
subj_to_obj_proper: 88.4
prim_to_obj_proper: 83.0
prim_to_subj_proper: 100.0
v_dat_p2_pp_moved_to_recipient: 13.109756097560975
LEX: 99.62666666666667
OVERALL: 89.27700675168792
INFO:root:Baselining the Transformer Encoder-Decoder Model
INFO:root:__Number CUDA Devices: 1
INFO:root:Number of model params: 4344077
INFO:root:OUTPUT DIR: ./results_cogs_train_default_wu_data_gen_on_modified_v_dat_p2/cogs_pipeline.model.ende_transformer.lf.cogs.glove.False.seed.67
Epoch: 0:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 0: 100% 213/213 [00:10<00:00, 20.26it/s, loss=6.01]
Epoch: 1:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 1: 100% 213/213 [00:10<00:00, 20.21it/s, loss=4.63]
Epoch: 2:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 2: 100% 213/213 [00:10<00:00, 20.32it/s, loss=3.63]
Epoch: 3:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 3: 100% 213/213 [00:10<00:00, 20.27it/s, loss=2.54]
Epoch: 4:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 4: 100% 213/213 [00:10<00:00, 20.33it/s, loss=1.94]
Epoch: 5:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 5: 100% 213/213 [00:10<00:00, 20.28it/s, loss=1.58]
Epoch: 6:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 6: 100% 213/213 [00:10<00:00, 20.27it/s, loss=1.32]
Epoch: 7:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 7: 100% 213/213 [00:10<00:00, 20.34it/s, loss=1.15]
Epoch: 8:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 8: 100% 213/213 [00:10<00:00, 20.36it/s, loss=1.03]
Epoch: 9:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 9: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.92]
Epoch: 10:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 10: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.83]
Epoch: 11:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 11: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.74]
Epoch: 12:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 12: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.67]
Epoch: 13:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 13: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.6]
Epoch: 14:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 14: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.52]
Epoch: 15:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 15: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0.46]
Epoch: 16:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 16: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.38]
Epoch: 17:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 17: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.34]
Epoch: 18:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 18: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.29]
Epoch: 19:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 19: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.26]
Epoch: 20:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 20: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.22]
Epoch: 21:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 21: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0.2]
Epoch: 22:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 22: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.18]
Epoch: 23:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 23: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.17]
Epoch: 24:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 24: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.14]
Epoch: 25:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 25: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.14]
Epoch: 26:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 26: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.12]
Epoch: 27:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 27: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0.1]
Epoch: 28:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 28: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.1]
Epoch: 29:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 29: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.1]
Epoch: 30:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 30: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.1]
Epoch: 31:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 31: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.08]
Epoch: 32:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 32: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.08]
Epoch: 33:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 33: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.07]
Epoch: 34:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 34: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.07]
Epoch: 35:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 35: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.06]
Epoch: 36:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 36: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.06]
Epoch: 37:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 37: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.06]
Epoch: 38:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 38: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.05]
Epoch: 39:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 39: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.05]
Epoch: 40:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 40: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.05]
Epoch: 41:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 41: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.04]
Epoch: 42:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 42: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.05]
Epoch: 43:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 43: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.04]
Epoch: 44:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 44: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0.04]
Epoch: 45:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 45: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.05]
Epoch: 46:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 46: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.04]
Epoch: 47:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 47: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.04]
Epoch: 48:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 48: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.04]
Epoch: 49:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 49: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.03]
Epoch: 50:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 50: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.03]
Epoch: 51:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 51: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.03]
Epoch: 52:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 52: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.03]
Epoch: 53:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 53: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.03]
Epoch: 54:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 54: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.03]
Epoch: 55:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 55: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.03]
Epoch: 56:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 56: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.03]
Epoch: 57:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 57: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.02]
Epoch: 58:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 58: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.03]
Epoch: 59:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 59: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.02]
Epoch: 60:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 60: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.03]
Epoch: 61:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 61: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.02]
Epoch: 62:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 62: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0.02]
Epoch: 63:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 63: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.02]
Epoch: 64:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 64: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.02]
Epoch: 65:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 65: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.02]
Epoch: 66:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 66: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.02]
Epoch: 67:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 67: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.02]
Epoch: 68:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 68: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.02]
Epoch: 69:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 69: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.02]
Epoch: 70:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 70: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 71:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 71: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.01]
Epoch: 72:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 72: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.02]
Epoch: 73:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 73: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.02]
Epoch: 74:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 74: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.01]
Epoch: 75:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 75: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.01]
Epoch: 76:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 76: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 77:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 77: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 78:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 78: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 79:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 79: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0.01]
Epoch: 80:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 80: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.01]
Epoch: 81:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 81: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.01]
Epoch: 82:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 82: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 83:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 83: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.01]
Epoch: 84:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 84: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 85:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 85: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.01]
Epoch: 86:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 86: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.01]
Epoch: 87:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 87: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.01]
Epoch: 88:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 88: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 89:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 89: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.01]
Epoch: 90:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 90: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 91:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 91: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.01]
Epoch: 92:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 92: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 93:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 93: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 94:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 94: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 95:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 95: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 96:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 96: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 97:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 97: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 98:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 98: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 99:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 99: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 100:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 100: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 101:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 101: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.01]
Epoch: 102:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 102: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.01]
Epoch: 103:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 103: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0.01]
Epoch: 104:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 104: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 105:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 105: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.01]
Epoch: 106:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 106: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 107:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 107: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 108:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 108: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 109:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 109: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 110:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 110: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 111:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 111: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 112:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 112: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 113:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 113: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 114:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 114: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 115:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 115: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 116:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 116: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 117:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 117: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 118:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 118: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 119:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 119: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 120:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 120: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 121:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 121: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 122:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 122: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 123:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 123: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.01]
Epoch: 124:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 124: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 125:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 125: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 126:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 126: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 127:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 127: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 128:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 128: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 129:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 129: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 130:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 130: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 131:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 131: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 132:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 132: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 133:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 133: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 134:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 134: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 135:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 135: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 136:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 136: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 137:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 137: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 138:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 138: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 139:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 139: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 140:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 140: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 141:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 141: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 142:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 142: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 143:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 143: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 144:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 144: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 145:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 145: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 146:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 146: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 147:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 147: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 148:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 148: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 149:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 149: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 150:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 150: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 151:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 151: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 152:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 152: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 153:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 153: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 154:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 154: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 155:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 155: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 156:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 156: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 157:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 157: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 158:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 158: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 159:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 159: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 160:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 160: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 161:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 161: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 162:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 162: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 163:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 163: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 164:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 164: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 165:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 165: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 166:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 166: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 167:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 167: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 168:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 168: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 169:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 169: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 170:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 170: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 171:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 171: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 172:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 172: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 173:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 173: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 174:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 174: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 175:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 175: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 176:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 176: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 177:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 177: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 178:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 178: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 179:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 179: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 180:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 180: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 181:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 181: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 182:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 182: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 183:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 183: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 184:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 184: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 185:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 185: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 186:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 186: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 187:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 187: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 188:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 188: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 189:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 189: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 190:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 190: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 191:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 191: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 192:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 192: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 193:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 193: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 194:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 194: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 195:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 195: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 196:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 196: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 197:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 197: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 198:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 198: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 199:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 199: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 200:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 200: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 201:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 201: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 202:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 202: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 203:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 203: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 204:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 204: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 205:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 205: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 206:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 206: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 207:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 207: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 208:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 208: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 209:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 209: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 210:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 210: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 211:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 211: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 212:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 212: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 213:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 213: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 214:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 214: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 215:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 215: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 216:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 216: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 217:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 217: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 218:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 218: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 219:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 219: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 220:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 220: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 221:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 221: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 222:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 222: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 223:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 223: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 224:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 224: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 225:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 225: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 226:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 226: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 227:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 227: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 228:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 228: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 229:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 229: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 230:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 230: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 231:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 231: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 232:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 232: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 233:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 233: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 234:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 234: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 235:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 235: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 236:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 236: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 237:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 237: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 238:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 238: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 239:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 239: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 240:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 240: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 241:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 241: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 242:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 242: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 243:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 243: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 244:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 244: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 245:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 245: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 246:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 246: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 247:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 247: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 248:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 248: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 249:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 249: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 250:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 250: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 251:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 251: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 252:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 252: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 253:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 253: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 254:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 254: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 255:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 255: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 256:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 256: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 257:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 257: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 258:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 258: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 259:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 259: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 260:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 260: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 261:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 261: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 262:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 262: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 263:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 263: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 264:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 264: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 265:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 265: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 266:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 266: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 267:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 267: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 268:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 268: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 269:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 269: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 270:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 270: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 271:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 271: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 272:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 272: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 273:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 273: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 274:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 274: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 275:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 275: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 276:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 276: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 277:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 277: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 278:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 278: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 279:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 279: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 280:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 280: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 281:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 281: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 282:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 282: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 283:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 283: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 284:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 284: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 285:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 285: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 286:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 286: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 287:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 287: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 288:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 288: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 289:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 289: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 290:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 290: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 291:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 291: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 292:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 292: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 293:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 293: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 294:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 294: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 295:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 295: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 296:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 296: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 297:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 297: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 298:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 298: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 299:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 299: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 100% 300/300 [52:39<00:00, 10.53s/it]
INFO:root:Training is finished ...
Iteration: 100% 24/24 [00:10<00:00,  2.27it/s, acc=1]
Iteration: 100% 167/167 [23:04<00:00,  8.29s/it, acc=0.866]
obj_pp_to_subj_pp: 18.1
cp_recursion: 52.8
pp_recursion: 32.0
subj_to_obj_proper: 82.8
prim_to_obj_proper: 63.4
prim_to_subj_proper: 100.0
v_dat_p2_pp_moved_to_recipient: 7.621951219512195
LEX: 99.73333333333333
OVERALL: 86.62790697674419
INFO:root:Baselining the Transformer Encoder-Decoder Model
INFO:root:__Number CUDA Devices: 1
INFO:root:Number of model params: 4344077
INFO:root:OUTPUT DIR: ./results_cogs_train_default_wu_data_gen_on_modified_v_dat_p2/cogs_pipeline.model.ende_transformer.lf.cogs.glove.False.seed.78
Epoch: 0:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 0: 100% 213/213 [00:10<00:00, 20.37it/s, loss=5.96]
Epoch: 1:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 1: 100% 213/213 [00:10<00:00, 20.40it/s, loss=4.65]
Epoch: 2:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 2: 100% 213/213 [00:10<00:00, 20.39it/s, loss=3.57]
Epoch: 3:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 3: 100% 213/213 [00:10<00:00, 20.31it/s, loss=2.5]
Epoch: 4:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 4: 100% 213/213 [00:10<00:00, 20.41it/s, loss=1.94]
Epoch: 5:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 5: 100% 213/213 [00:10<00:00, 20.40it/s, loss=1.61]
Epoch: 6:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 6: 100% 213/213 [00:10<00:00, 20.42it/s, loss=1.35]
Epoch: 7:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 7: 100% 213/213 [00:10<00:00, 20.41it/s, loss=1.17]
Epoch: 8:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 8: 100% 213/213 [00:10<00:00, 20.35it/s, loss=1.03]
Epoch: 9:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 9: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.91]
Epoch: 10:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 10: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.82]
Epoch: 11:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 11: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.72]
Epoch: 12:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 12: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.63]
Epoch: 13:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 13: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.55]
Epoch: 14:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 14: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0.48]
Epoch: 15:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 15: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.42]
Epoch: 16:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 16: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0.36]
Epoch: 17:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 17: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0.32]
Epoch: 18:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 18: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.27]
Epoch: 19:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 19: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.24]
Epoch: 20:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 20: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.21]
Epoch: 21:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 21: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.17]
Epoch: 22:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 22: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.16]
Epoch: 23:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 23: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.14]
Epoch: 24:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 24: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.14]
Epoch: 25:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 25: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.12]
Epoch: 26:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 26: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0.11]
Epoch: 27:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 27: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.09]
Epoch: 28:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 28: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.09]
Epoch: 29:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 29: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0.07]
Epoch: 30:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 30: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.07]
Epoch: 31:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 31: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.06]
Epoch: 32:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 32: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.06]
Epoch: 33:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 33: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.05]
Epoch: 34:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 34: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.05]
Epoch: 35:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 35: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.04]
Epoch: 36:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 36: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.04]
Epoch: 37:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 37: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.04]
Epoch: 38:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 38: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.03]
Epoch: 39:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 39: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.03]
Epoch: 40:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 40: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.02]
Epoch: 41:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 41: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.03]
Epoch: 42:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 42: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.02]
Epoch: 43:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 43: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.02]
Epoch: 44:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 44: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0.02]
Epoch: 45:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 45: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.02]
Epoch: 46:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 46: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.02]
Epoch: 47:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 47: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.01]
Epoch: 48:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 48: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.01]
Epoch: 49:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 49: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.01]
Epoch: 50:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 50: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.01]
Epoch: 51:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 51: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.01]
Epoch: 52:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 52: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.02]
Epoch: 53:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 53: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.01]
Epoch: 54:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 54: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.01]
Epoch: 55:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 55: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.01]
Epoch: 56:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 56: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.01]
Epoch: 57:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 57: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 58:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 58: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.01]
Epoch: 59:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 59: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.01]
Epoch: 60:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 60: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.01]
Epoch: 61:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 61: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0.01]
Epoch: 62:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 62: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.01]
Epoch: 63:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 63: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.01]
Epoch: 64:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 64: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.01]
Epoch: 65:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 65: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 66:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 66: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 67:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 67: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 68:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 68: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 69:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 69: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.01]
Epoch: 70:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 70: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.01]
Epoch: 71:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 71: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 72:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 72: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 73:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 73: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 74:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 74: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 75:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 75: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 76:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 76: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 77:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 77: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 78:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 78: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 79:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 79: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0.01]
Epoch: 80:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 80: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.01]
Epoch: 81:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 81: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 82:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 82: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 83:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 83: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 84:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 84: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 85:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 85: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 86:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 86: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 87:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 87: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 88:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 88: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 89:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 89: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 90:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 90: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 91:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 91: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 92:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 92: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 93:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 93: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 94:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 94: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 95:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 95: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 96:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 96: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 97:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 97: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 98:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 98: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 99:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 99: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 100:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 100: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 101:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 101: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 102:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 102: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 103:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 103: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 104:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 104: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 105:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 105: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 106:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 106: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 107:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 107: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 108:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 108: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 109:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 109: 100% 213/213 [00:10<00:00, 20.08it/s, loss=0]
Epoch: 110:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 110: 100% 213/213 [00:10<00:00, 20.09it/s, loss=0]
Epoch: 111:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 111: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 112:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 112: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 113:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 113: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 114:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 114: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 115:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 115: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 116:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 116: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.01]
Epoch: 117:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 117: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 118:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 118: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 119:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 119: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0.01]
Epoch: 120:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 120: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 121:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 121: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 122:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 122: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 123:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 123: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 124:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 124: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 125:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 125: 100% 213/213 [00:10<00:00, 20.10it/s, loss=0]
Epoch: 126:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 126: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 127:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 127: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 128:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 128: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 129:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 129: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 130:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 130: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 131:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 131: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 132:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 132: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 133:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 133: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 134:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 134: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 135:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 135: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 136:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 136: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 137:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 137: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 138:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 138: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 139:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 139: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 140:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 140: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 141:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 141: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 142:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 142: 100% 213/213 [00:10<00:00, 20.06it/s, loss=0]
Epoch: 143:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 143: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 144:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 144: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 145:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 145: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 146:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 146: 100% 213/213 [00:10<00:00, 20.07it/s, loss=0]
Epoch: 147:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 147: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 148:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 148: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 149:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 149: 100% 213/213 [00:10<00:00, 20.08it/s, loss=0]
Epoch: 150:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 150: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 151:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 151: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 152:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 152: 100% 213/213 [00:10<00:00, 20.09it/s, loss=0]
Epoch: 153:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 153: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 154:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 154: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 155:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 155: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 156:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 156: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 157:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 157: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 158:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 158: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 159:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 159: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 160:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 160: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 161:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 161: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 162:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 162: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 163:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 163: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 164:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 164: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 165:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 165: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 166:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 166: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 167:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 167: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 168:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 168: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 169:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 169: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 170:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 170: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 171:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 171: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 172:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 172: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 173:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 173: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 174:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 174: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 175:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 175: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 176:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 176: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 177:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 177: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 178:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 178: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 179:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 179: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 180:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 180: 100% 213/213 [00:10<00:00, 20.03it/s, loss=0]
Epoch: 181:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 181: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 182:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 182: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 183:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 183: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 184:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 184: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 185:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 185: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 186:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 186: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 187:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 187: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 188:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 188: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 189:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 189: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 190:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 190: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 191:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 191: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 192:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 192: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 193:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 193: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 194:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 194: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 195:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 195: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 196:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 196: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 197:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 197: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 198:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 198: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 199:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 199: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 200:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 200: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 201:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 201: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 202:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 202: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 203:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 203: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 204:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 204: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 205:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 205: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 206:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 206: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 207:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 207: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 208:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 208: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 209:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 209: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 210:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 210: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 211:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 211: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 212:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 212: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 213:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 213: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 214:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 214: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 215:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 215: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 216:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 216: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 217:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 217: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0]
Epoch: 218:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 218: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 219:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 219: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 220:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 220: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 221:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 221: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 222:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 222: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 223:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 223: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0]
Epoch: 224:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 224: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 225:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 225: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 226:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 226: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 227:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 227: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 228:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 228: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 229:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 229: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 230:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 230: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 231:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 231: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 232:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 232: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 233:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 233: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 234:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 234: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 235:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 235: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 236:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 236: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 237:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 237: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 238:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 238: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 239:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 239: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 240:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 240: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 241:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 241: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 242:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 242: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 243:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 243: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 244:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 244: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 245:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 245: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 246:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 246: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 247:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 247: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 248:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 248: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 249:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 249: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 250:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 250: 100% 213/213 [00:10<00:00, 20.10it/s, loss=0]
Epoch: 251:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 251: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 252:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 252: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 253:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 253: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 254:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 254: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 255:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 255: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 256:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 256: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 257:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 257: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 258:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 258: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 259:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 259: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 260:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 260: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 261:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 261: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 262:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 262: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 263:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 263: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 264:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 264: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 265:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 265: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 266:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 266: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 267:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 267: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 268:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 268: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 269:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 269: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 270:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 270: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 271:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 271: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 272:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 272: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 273:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 273: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 274:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 274: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 275:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 275: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 276:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 276: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 277:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 277: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 278:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 278: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 279:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 279: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 280:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 280: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 281:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 281: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 282:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 282: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 283:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 283: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 284:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 284: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 285:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 285: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 286:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 286: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 287:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 287: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 288:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 288: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 289:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 289: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 290:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 290: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 291:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 291: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 292:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 292: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 293:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 293: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 294:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 294: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 295:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 295: 100% 213/213 [00:10<00:00, 20.07it/s, loss=0]
Epoch: 296:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 296: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 297:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 297: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 298:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 298: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 299:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 299: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 100% 300/300 [52:41<00:00, 10.54s/it]
INFO:root:Training is finished ...
Iteration: 100% 24/24 [00:10<00:00,  2.24it/s, acc=1]
Iteration: 100% 167/167 [17:32<00:00,  6.30s/it, acc=0.862]
obj_pp_to_subj_pp: 20.0
cp_recursion: 49.3
pp_recursion: 50.1
subj_to_obj_proper: 94.7
prim_to_obj_proper: 88.8
prim_to_subj_proper: 100.0
v_dat_p2_pp_moved_to_recipient: 0.6097560975609756
LEX: 95.64666666666668
OVERALL: 86.16841710427607
INFO:root:Baselining the Transformer Encoder-Decoder Model
INFO:root:__Number CUDA Devices: 1
INFO:root:Number of model params: 4344077
INFO:root:OUTPUT DIR: ./results_cogs_train_default_wu_data_gen_on_modified_v_dat_p2/cogs_pipeline.model.ende_transformer.lf.cogs.glove.False.seed.89
Epoch: 0:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 0: 100% 213/213 [00:10<00:00, 20.33it/s, loss=6.03]
Epoch: 1:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 1: 100% 213/213 [00:10<00:00, 20.43it/s, loss=4.65]
Epoch: 2:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 2: 100% 213/213 [00:10<00:00, 20.41it/s, loss=3.6]
Epoch: 3:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 3: 100% 213/213 [00:10<00:00, 20.43it/s, loss=2.53]
Epoch: 4:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 4: 100% 213/213 [00:10<00:00, 20.39it/s, loss=1.97]
Epoch: 5:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 5: 100% 213/213 [00:10<00:00, 20.43it/s, loss=1.61]
Epoch: 6:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 6: 100% 213/213 [00:10<00:00, 20.46it/s, loss=1.34]
Epoch: 7:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 7: 100% 213/213 [00:10<00:00, 20.46it/s, loss=1.16]
Epoch: 8:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 8: 100% 213/213 [00:10<00:00, 20.45it/s, loss=1.03]
Epoch: 9:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 9: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0.93]
Epoch: 10:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 10: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.84]
Epoch: 11:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 11: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0.74]
Epoch: 12:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 12: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0.66]
Epoch: 13:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 13: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0.58]
Epoch: 14:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 14: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0.5]
Epoch: 15:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 15: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0.42]
Epoch: 16:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 16: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.37]
Epoch: 17:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 17: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0.31]
Epoch: 18:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 18: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0.26]
Epoch: 19:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 19: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.23]
Epoch: 20:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 20: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.19]
Epoch: 21:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 21: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.16]
Epoch: 22:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 22: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0.15]
Epoch: 23:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 23: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0.12]
Epoch: 24:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 24: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0.11]
Epoch: 25:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 25: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.1]
Epoch: 26:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 26: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0.09]
Epoch: 27:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 27: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.09]
Epoch: 28:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 28: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.07]
Epoch: 29:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 29: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.06]
Epoch: 30:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 30: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.06]
Epoch: 31:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 31: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0.05]
Epoch: 32:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 32: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.06]
Epoch: 33:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 33: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0.05]
Epoch: 34:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 34: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.05]
Epoch: 35:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 35: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.04]
Epoch: 36:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 36: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0.04]
Epoch: 37:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 37: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.04]
Epoch: 38:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 38: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.03]
Epoch: 39:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 39: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.03]
Epoch: 40:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 40: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.03]
Epoch: 41:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 41: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.03]
Epoch: 42:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 42: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0.03]
Epoch: 43:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 43: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.02]
Epoch: 44:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 44: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.02]
Epoch: 45:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 45: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.02]
Epoch: 46:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 46: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.03]
Epoch: 47:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 47: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0.02]
Epoch: 48:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 48: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.02]
Epoch: 49:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 49: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0.02]
Epoch: 50:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 50: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0.02]
Epoch: 51:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 51: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.02]
Epoch: 52:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 52: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0.02]
Epoch: 53:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 53: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.01]
Epoch: 54:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 54: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.02]
Epoch: 55:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 55: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0.01]
Epoch: 56:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 56: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.01]
Epoch: 57:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 57: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 58:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 58: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.01]
Epoch: 59:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 59: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.02]
Epoch: 60:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 60: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.01]
Epoch: 61:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 61: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.01]
Epoch: 62:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 62: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.01]
Epoch: 63:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 63: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0.01]
Epoch: 64:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 64: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.01]
Epoch: 65:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 65: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0.01]
Epoch: 66:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 66: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0.01]
Epoch: 67:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 67: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0.01]
Epoch: 68:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 68: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 69:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 69: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0.01]
Epoch: 70:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 70: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.01]
Epoch: 71:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 71: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.01]
Epoch: 72:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 72: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0.01]
Epoch: 73:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 73: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.01]
Epoch: 74:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 74: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.01]
Epoch: 75:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 75: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.01]
Epoch: 76:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 76: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.01]
Epoch: 77:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 77: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.01]
Epoch: 78:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 78: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.01]
Epoch: 79:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 79: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 80:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 80: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 81:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 81: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 82:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 82: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.01]
Epoch: 83:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 83: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.01]
Epoch: 84:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 84: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.01]
Epoch: 85:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 85: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0]
Epoch: 86:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 86: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.01]
Epoch: 87:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 87: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.01]
Epoch: 88:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 88: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0.01]
Epoch: 89:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 89: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0.01]
Epoch: 90:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 90: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0.01]
Epoch: 91:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 91: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 92:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 92: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0.01]
Epoch: 93:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 93: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 94:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 94: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 95:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 95: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 96:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 96: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 97:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 97: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 98:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 98: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.01]
Epoch: 99:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 99: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 100:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 100: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0.01]
Epoch: 101:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 101: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 102:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 102: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 103:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 103: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0.01]
Epoch: 104:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 104: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 105:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 105: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 106:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 106: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 107:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 107: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 108:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 108: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 109:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 109: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0.01]
Epoch: 110:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 110: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 111:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 111: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 112:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 112: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 113:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 113: 100% 213/213 [00:10<00:00, 20.48it/s, loss=0]
Epoch: 114:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 114: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 115:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 115: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 116:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 116: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 117:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 117: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 118:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 118: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 119:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 119: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 120:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 120: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0.01]
Epoch: 121:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 121: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 122:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 122: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 123:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 123: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 124:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 124: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 125:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 125: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 126:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 126: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 127:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 127: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 128:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 128: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 129:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 129: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 130:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 130: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 131:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 131: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 132:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 132: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 133:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 133: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0.01]
Epoch: 134:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 134: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 135:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 135: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 136:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 136: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 137:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 137: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 138:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 138: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 139:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 139: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0]
Epoch: 140:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 140: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 141:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 141: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 142:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 142: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 143:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 143: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 144:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 144: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 145:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 145: 100% 213/213 [00:10<00:00, 20.48it/s, loss=0]
Epoch: 146:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 146: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 147:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 147: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 148:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 148: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 149:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 149: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 150:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 150: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 151:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 151: 100% 213/213 [00:10<00:00, 20.48it/s, loss=0]
Epoch: 152:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 152: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 153:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 153: 100% 213/213 [00:10<00:00, 20.48it/s, loss=0]
Epoch: 154:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 154: 100% 213/213 [00:10<00:00, 20.49it/s, loss=0]
Epoch: 155:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 155: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 156:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 156: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 157:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 157: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 158:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 158: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 159:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 159: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 160:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 160: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 161:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 161: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 162:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 162: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 163:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 163: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 164:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 164: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 165:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 165: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 166:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 166: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 167:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 167: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 168:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 168: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 169:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 169: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 170:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 170: 100% 213/213 [00:10<00:00, 20.48it/s, loss=0]
Epoch: 171:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 171: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0]
Epoch: 172:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 172: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 173:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 173: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 174:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 174: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 175:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 175: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 176:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 176: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0]
Epoch: 177:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 177: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 178:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 178: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 179:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 179: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 180:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 180: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 181:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 181: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 182:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 182: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 183:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 183: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 184:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 184: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 185:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 185: 100% 213/213 [00:10<00:00, 20.14it/s, loss=0]
Epoch: 186:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 186: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 187:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 187: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 188:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 188: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 189:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 189: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 190:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 190: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0]
Epoch: 191:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 191: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 192:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 192: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 193:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 193: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 194:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 194: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 195:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 195: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 196:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 196: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 197:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 197: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 198:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 198: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 199:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 199: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 200:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 200: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 201:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 201: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 202:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 202: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 203:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 203: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 204:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 204: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 205:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 205: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 206:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 206: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 207:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 207: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 208:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 208: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 209:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 209: 100% 213/213 [00:10<00:00, 20.17it/s, loss=0]
Epoch: 210:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 210: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 211:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 211: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 212:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 212: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 213:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 213: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 214:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 214: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 215:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 215: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 216:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 216: 100% 213/213 [00:10<00:00, 19.91it/s, loss=0]
Epoch: 217:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 217: 100% 213/213 [00:10<00:00, 19.88it/s, loss=0]
Epoch: 218:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 218: 100% 213/213 [00:10<00:00, 20.04it/s, loss=0]
Epoch: 219:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 219: 100% 213/213 [00:10<00:00, 19.90it/s, loss=0]
Epoch: 220:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 220: 100% 213/213 [00:10<00:00, 19.81it/s, loss=0]
Epoch: 221:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 221: 100% 213/213 [00:10<00:00, 19.99it/s, loss=0]
Epoch: 222:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 222: 100% 213/213 [00:10<00:00, 19.84it/s, loss=0]
Epoch: 223:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 223: 100% 213/213 [00:10<00:00, 19.88it/s, loss=0]
Epoch: 224:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 224: 100% 213/213 [00:10<00:00, 19.92it/s, loss=0]
Epoch: 225:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 225: 100% 213/213 [00:10<00:00, 19.88it/s, loss=0]
Epoch: 226:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 226: 100% 213/213 [00:10<00:00, 19.91it/s, loss=0]
Epoch: 227:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 227: 100% 213/213 [00:10<00:00, 19.85it/s, loss=0]
Epoch: 228:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 228: 100% 213/213 [00:10<00:00, 19.88it/s, loss=0]
Epoch: 229:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 229: 100% 213/213 [00:10<00:00, 19.97it/s, loss=0]
Epoch: 230:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 230: 100% 213/213 [00:10<00:00, 19.93it/s, loss=0]
Epoch: 231:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 231: 100% 213/213 [00:10<00:00, 19.98it/s, loss=0]
Epoch: 232:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 232: 100% 213/213 [00:10<00:00, 19.96it/s, loss=0]
Epoch: 233:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 233: 100% 213/213 [00:10<00:00, 19.96it/s, loss=0]
Epoch: 234:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 234: 100% 213/213 [00:10<00:00, 20.03it/s, loss=0]
Epoch: 235:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 235: 100% 213/213 [00:10<00:00, 19.96it/s, loss=0]
Epoch: 236:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 236: 100% 213/213 [00:10<00:00, 19.99it/s, loss=0]
Epoch: 237:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 237: 100% 213/213 [00:10<00:00, 20.01it/s, loss=0]
Epoch: 238:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 238: 100% 213/213 [00:10<00:00, 19.95it/s, loss=0]
Epoch: 239:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 239: 100% 213/213 [00:10<00:00, 20.00it/s, loss=0]
Epoch: 240:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 240: 100% 213/213 [00:10<00:00, 19.98it/s, loss=0]
Epoch: 241:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 241: 100% 213/213 [00:10<00:00, 19.93it/s, loss=0]
Epoch: 242:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 242: 100% 213/213 [00:10<00:00, 20.03it/s, loss=0]
Epoch: 243:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 243: 100% 213/213 [00:10<00:00, 19.87it/s, loss=0]
Epoch: 244:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 244: 100% 213/213 [00:10<00:00, 19.99it/s, loss=0]
Epoch: 245:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 245: 100% 213/213 [00:10<00:00, 20.01it/s, loss=0]
Epoch: 246:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 246: 100% 213/213 [00:10<00:00, 19.93it/s, loss=0]
Epoch: 247:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 247: 100% 213/213 [00:10<00:00, 20.01it/s, loss=0]
Epoch: 248:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 248: 100% 213/213 [00:10<00:00, 19.92it/s, loss=0]
Epoch: 249:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 249: 100% 213/213 [00:10<00:00, 19.87it/s, loss=0]
Epoch: 250:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 250: 100% 213/213 [00:10<00:00, 20.01it/s, loss=0]
Epoch: 251:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 251: 100% 213/213 [00:10<00:00, 19.93it/s, loss=0]
Epoch: 252:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 252: 100% 213/213 [00:10<00:00, 19.94it/s, loss=0]
Epoch: 253:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 253: 100% 213/213 [00:10<00:00, 19.93it/s, loss=0]
Epoch: 254:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 254: 100% 213/213 [00:10<00:00, 19.91it/s, loss=0]
Epoch: 255:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 255: 100% 213/213 [00:10<00:00, 20.00it/s, loss=0]
Epoch: 256:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 256: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 257:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 257: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 258:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 258: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 259:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 259: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 260:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 260: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 261:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 261: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 262:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 262: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 263:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 263: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 264:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 264: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 265:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 265: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 266:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 266: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 267:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 267: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 268:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 268: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0]
Epoch: 269:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 269: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 270:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 270: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 271:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 271: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 272:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 272: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 273:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 273: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0]
Epoch: 274:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 274: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 275:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 275: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 276:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 276: 100% 213/213 [00:10<00:00, 20.47it/s, loss=0]
Epoch: 277:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 277: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 278:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 278: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 279:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 279: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 280:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 280: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 281:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 281: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 282:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 282: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 283:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 283: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 284:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 284: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 285:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 285: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 286:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 286: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 287:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 287: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0]
Epoch: 288:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 288: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 289:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 289: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 290:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 290: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 291:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 291: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 292:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 292: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 293:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 293: 100% 213/213 [00:10<00:00, 20.44it/s, loss=0]
Epoch: 294:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 294: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 295:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 295: 100% 213/213 [00:10<00:00, 20.42it/s, loss=0]
Epoch: 296:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 296: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 297:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 297: 100% 213/213 [00:10<00:00, 20.45it/s, loss=0]
Epoch: 298:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 298: 100% 213/213 [00:10<00:00, 20.43it/s, loss=0]
Epoch: 299:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 299: 100% 213/213 [00:10<00:00, 20.46it/s, loss=0]
Epoch: 100% 300/300 [52:38<00:00, 10.53s/it]
INFO:root:Training is finished ...
Iteration: 100% 24/24 [00:10<00:00,  2.25it/s, acc=1]
Iteration: 100% 167/167 [17:10<00:00,  6.17s/it, acc=0.883]
obj_pp_to_subj_pp: 20.2
cp_recursion: 53.5
pp_recursion: 32.0
subj_to_obj_proper: 87.5
prim_to_obj_proper: 97.0
prim_to_subj_proper: 100.0
v_dat_p2_pp_moved_to_recipient: 0.0
LEX: 99.52
OVERALL: 88.28769692423106
INFO:root:Baselining the Transformer Encoder-Decoder Model
INFO:root:__Number CUDA Devices: 1
INFO:root:Number of model params: 4344077
INFO:root:OUTPUT DIR: ./results_cogs_train_default_wu_data_gen_on_modified_v_dat_p2/cogs_pipeline.model.ende_transformer.lf.cogs.glove.False.seed.100
Epoch: 0:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 0: 100% 213/213 [00:10<00:00, 20.27it/s, loss=6.01]
Epoch: 1:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 1: 100% 213/213 [00:10<00:00, 20.32it/s, loss=4.57]
Epoch: 2:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 2: 100% 213/213 [00:10<00:00, 20.27it/s, loss=3.56]
Epoch: 3:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 3: 100% 213/213 [00:10<00:00, 20.32it/s, loss=2.5]
Epoch: 4:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 4: 100% 213/213 [00:10<00:00, 20.32it/s, loss=1.92]
Epoch: 5:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 5: 100% 213/213 [00:10<00:00, 20.33it/s, loss=1.58]
Epoch: 6:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 6: 100% 213/213 [00:10<00:00, 20.33it/s, loss=1.32]
Epoch: 7:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 7: 100% 213/213 [00:10<00:00, 20.24it/s, loss=1.15]
Epoch: 8:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 8: 100% 213/213 [00:10<00:00, 20.33it/s, loss=1.02]
Epoch: 9:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 9: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.92]
Epoch: 10:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 10: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.82]
Epoch: 11:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 11: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.72]
Epoch: 12:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 12: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0.64]
Epoch: 13:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 13: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.56]
Epoch: 14:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 14: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.49]
Epoch: 15:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 15: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0.43]
Epoch: 16:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 16: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.37]
Epoch: 17:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 17: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.31]
Epoch: 18:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 18: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0.26]
Epoch: 19:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 19: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0.22]
Epoch: 20:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 20: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0.17]
Epoch: 21:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 21: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0.15]
Epoch: 22:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 22: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.12]
Epoch: 23:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 23: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.1]
Epoch: 24:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 24: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.09]
Epoch: 25:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 25: 100% 213/213 [00:10<00:00, 20.09it/s, loss=0.07]
Epoch: 26:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 26: 100% 213/213 [00:10<00:00, 20.01it/s, loss=0.06]
Epoch: 27:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 27: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0.06]
Epoch: 28:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 28: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0.05]
Epoch: 29:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 29: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.05]
Epoch: 30:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 30: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.04]
Epoch: 31:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 31: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.06]
Epoch: 32:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 32: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0.04]
Epoch: 33:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 33: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0.03]
Epoch: 34:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 34: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.04]
Epoch: 35:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 35: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0.02]
Epoch: 36:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 36: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0.03]
Epoch: 37:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 37: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0.02]
Epoch: 38:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 38: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0.01]
Epoch: 39:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 39: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.02]
Epoch: 40:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 40: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0.02]
Epoch: 41:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 41: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.01]
Epoch: 42:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 42: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0.01]
Epoch: 43:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 43: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.02]
Epoch: 44:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 44: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0.01]
Epoch: 45:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 45: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0.01]
Epoch: 46:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 46: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.01]
Epoch: 47:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 47: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.01]
Epoch: 48:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 48: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0.01]
Epoch: 49:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 49: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0.01]
Epoch: 50:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 50: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0.01]
Epoch: 51:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 51: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.01]
Epoch: 52:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 52: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0.01]
Epoch: 53:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 53: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0.01]
Epoch: 54:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 54: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 55:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 55: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0.01]
Epoch: 56:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 56: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.01]
Epoch: 57:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 57: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0.01]
Epoch: 58:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 58: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0.01]
Epoch: 59:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 59: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 60:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 60: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.01]
Epoch: 61:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 61: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 62:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 62: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 63:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 63: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 64:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 64: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.01]
Epoch: 65:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 65: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 66:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 66: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0.01]
Epoch: 67:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 67: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.01]
Epoch: 68:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 68: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 69:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 69: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 70:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 70: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 71:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 71: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 72:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 72: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 73:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 73: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 74:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 74: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 75:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 75: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 76:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 76: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 77:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 77: 100% 213/213 [00:10<00:00, 20.12it/s, loss=0]
Epoch: 78:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 78: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 79:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 79: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 80:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 80: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 81:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 81: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 82:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 82: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 83:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 83: 100% 213/213 [00:10<00:00, 20.13it/s, loss=0]
Epoch: 84:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 84: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 85:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 85: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 86:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 86: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 87:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 87: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0.01]
Epoch: 88:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 88: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 89:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 89: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 90:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 90: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 91:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 91: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 92:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 92: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 93:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 93: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 94:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 94: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 95:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 95: 100% 213/213 [00:10<00:00, 20.21it/s, loss=0]
Epoch: 96:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 96: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 97:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 97: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 98:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 98: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 99:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 99: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 100:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 100: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 101:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 101: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 102:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 102: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 103:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 103: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 104:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 104: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 105:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 105: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 106:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 106: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 107:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 107: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 108:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 108: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 109:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 109: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 110:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 110: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 111:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 111: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 112:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 112: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 113:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 113: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 114:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 114: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 115:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 115: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 116:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 116: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 117:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 117: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 118:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 118: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 119:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 119: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 120:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 120: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 121:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 121: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 122:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 122: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 123:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 123: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 124:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 124: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 125:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 125: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 126:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 126: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 127:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 127: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 128:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 128: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 129:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 129: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 130:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 130: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 131:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 131: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 132:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 132: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 133:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 133: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 134:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 134: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 135:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 135: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 136:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 136: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 137:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 137: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 138:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 138: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 139:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 139: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 140:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 140: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 141:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 141: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 142:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 142: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 143:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 143: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 144:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 144: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 145:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 145: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 146:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 146: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 147:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 147: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 148:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 148: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 149:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 149: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 150:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 150: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 151:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 151: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 152:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 152: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 153:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 153: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 154:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 154: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 155:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 155: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 156:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 156: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 157:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 157: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 158:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 158: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 159:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 159: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 160:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 160: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 161:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 161: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 162:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 162: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 163:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 163: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 164:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 164: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 165:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 165: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 166:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 166: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 167:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 167: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 168:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 168: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 169:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 169: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 170:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 170: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 171:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 171: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 172:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 172: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 173:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 173: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 174:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 174: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 175:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 175: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 176:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 176: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 177:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 177: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 178:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 178: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 179:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 179: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 180:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 180: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 181:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 181: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 182:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 182: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 183:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 183: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 184:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 184: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 185:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 185: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 186:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 186: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 187:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 187: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 188:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 188: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 189:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 189: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 190:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 190: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 191:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 191: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 192:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 192: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 193:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 193: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 194:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 194: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 195:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 195: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 196:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 196: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 197:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 197: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 198:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 198: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 199:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 199: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 200:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 200: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 201:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 201: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 202:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 202: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 203:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 203: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 204:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 204: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 205:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 205: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 206:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 206: 100% 213/213 [00:10<00:00, 20.39it/s, loss=0]
Epoch: 207:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 207: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 208:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 208: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 209:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 209: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 210:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 210: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 211:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 211: 100% 213/213 [00:10<00:00, 20.20it/s, loss=0]
Epoch: 212:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 212: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 213:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 213: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 214:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 214: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 215:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 215: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 216:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 216: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 217:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 217: 100% 213/213 [00:10<00:00, 20.28it/s, loss=0]
Epoch: 218:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 218: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 219:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 219: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 220:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 220: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 221:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 221: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 222:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 222: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 223:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 223: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 224:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 224: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 225:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 225: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 226:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 226: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 227:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 227: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 228:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 228: 100% 213/213 [00:10<00:00, 20.25it/s, loss=0]
Epoch: 229:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 229: 100% 213/213 [00:10<00:00, 20.18it/s, loss=0]
Epoch: 230:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 230: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 231:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 231: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 232:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 232: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 233:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 233: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 234:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 234: 100% 213/213 [00:10<00:00, 20.15it/s, loss=0]
Epoch: 235:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 235: 100% 213/213 [00:10<00:00, 20.23it/s, loss=0]
Epoch: 236:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 236: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 237:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 237: 100% 213/213 [00:10<00:00, 20.24it/s, loss=0]
Epoch: 238:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 238: 100% 213/213 [00:10<00:00, 20.26it/s, loss=0]
Epoch: 239:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 239: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 240:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 240: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 241:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 241: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 242:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 242: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 243:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 243: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 244:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 244: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 245:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 245: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 246:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 246: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 247:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 247: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 248:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 248: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 249:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 249: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 250:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 250: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 251:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 251: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 252:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 252: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 253:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 253: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 254:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 254: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 255:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 255: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 256:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 256: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 257:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 257: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 258:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 258: 100% 213/213 [00:10<00:00, 20.27it/s, loss=0]
Epoch: 259:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 259: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 260:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 260: 100% 213/213 [00:10<00:00, 20.34it/s, loss=0]
Epoch: 261:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 261: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 262:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 262: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 263:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 263: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 264:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 264: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 265:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 265: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 266:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 266: 100% 213/213 [00:10<00:00, 20.19it/s, loss=0]
Epoch: 267:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 267: 100% 213/213 [00:10<00:00, 20.16it/s, loss=0]
Epoch: 268:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 268: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 269:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 269: 100% 213/213 [00:10<00:00, 20.22it/s, loss=0]
Epoch: 270:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 270: 100% 213/213 [00:10<00:00, 20.32it/s, loss=0]
Epoch: 271:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 271: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 272:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 272: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 273:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 273: 100% 213/213 [00:10<00:00, 20.40it/s, loss=0]
Epoch: 274:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 274: 100% 213/213 [00:10<00:00, 20.41it/s, loss=0]
Epoch: 275:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 275: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 276:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 276: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 277:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 277: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 278:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 278: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 279:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 279: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 280:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 280: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 281:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 281: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 282:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 282: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 283:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 283: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 284:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 284: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 285:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 285: 100% 213/213 [00:10<00:00, 20.33it/s, loss=0]
Epoch: 286:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 286: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 287:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 287: 100% 213/213 [00:10<00:00, 20.29it/s, loss=0]
Epoch: 288:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 288: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 289:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 289: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 290:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 290: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 291:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 291: 100% 213/213 [00:10<00:00, 20.35it/s, loss=0]
Epoch: 292:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 292: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 293:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 293: 100% 213/213 [00:10<00:00, 20.31it/s, loss=0]
Epoch: 294:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 294: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 295:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 295: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 296:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 296: 100% 213/213 [00:10<00:00, 20.36it/s, loss=0]
Epoch: 297:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 297: 100% 213/213 [00:10<00:00, 20.38it/s, loss=0]
Epoch: 298:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 298: 100% 213/213 [00:10<00:00, 20.37it/s, loss=0]
Epoch: 299:   0% 0/213 [00:00<?, ?it/s]/content/ReCOGS/model/encoder_decoder_hf.py:828: FutureWarning: Version v4.12.0 introduces a better way to train encoder-decoder models by computing the loss inside the encoder-decoder framework rather than in the decoder itself. You may observe training discrepancies if fine-tuning a model trained with versions anterior to 4.12.0. The decoder_input_ids are now created based on the labels, no need to pass them yourself anymore.
  warnings.warn(DEPRECATION_WARNING, FutureWarning)
Epoch: 299: 100% 213/213 [00:10<00:00, 20.30it/s, loss=0]
Epoch: 100% 300/300 [52:44<00:00, 10.55s/it]
INFO:root:Training is finished ...
Iteration: 100% 24/24 [00:10<00:00,  2.25it/s, acc=1]
Iteration: 100% 167/167 [18:57<00:00,  6.81s/it, acc=0.877]
obj_pp_to_subj_pp: 16.3
cp_recursion: 53.2
pp_recursion: 30.5
subj_to_obj_proper: 97.4
prim_to_obj_proper: 81.2
prim_to_subj_proper: 99.9
v_dat_p2_pp_moved_to_recipient: 43.292682926829265
LEX: 98.58666666666667
OVERALL: 87.74849962490623
In [ ]:
import numpy as np
wu_et_al_2023_baseline_v_dat_p2_pp_moved_to_recipient_cogs_pos = np.array([0.0, 9.45,11.58, 4.87, 43.29, 13.109756097560975, 7.621951219512195, 0.6097560975609756, 0.0,43.292682926829265 ])
In [ ]:
wu_et_al_2023_baseline_v_dat_p2_pp_moved_to_recipient_cogs_pos.mean()
Out[ ]:
13.38241463414634
In [ ]:
wu_et_al_2023_baseline_v_dat_p2_pp_moved_to_recipient_cogs_pos.std()
Out[ ]:
15.60396657401414
In [ ]:
import math
wu_et_al_2023_baseline_v_dat_p2_pp_moved_to_recipient_cogs_pos.std()/math.sqrt(len(wu_et_al_2023_baseline_v_dat_p2_pp_moved_to_recipient_cogs_pos))
Out[ ]:
4.934407490701903
In [ ]:
stderr_1p96 = wu_et_al_2023_baseline_v_dat_p2_pp_moved_to_recipient_cogs_pos.std()/math.sqrt(len(wu_et_al_2023_baseline_v_dat_p2_pp_moved_to_recipient_cogs_pos))*1.96
In [ ]:
(wu_et_al_2023_baseline_v_dat_p2_pp_moved_to_recipient_cogs_pos.mean() - stderr_1p96, wu_et_al_2023_baseline_v_dat_p2_pp_moved_to_recipient_cogs_pos.mean() + stderr_1p96)
Out[ ]:
(3.7109759523706085, 23.05385331592207)
In [ ]: